Xpress Release Notes

New in Release 9.5

Xpress Optimizer 44.01.01

• New functions XPRSgetcallbacksolution, XPRSgetcallbackslacks, XPRSgetcallbackduals, and XPRSgetcallbackredcosts. These functions can be used for retrieving the context-specific solution for any problem type and any callback where a solution is available. The solution returned in the affected callbacks are:
  • bariteration: Barrier iteration solution
  • chgbranchobject, nodelpsolved, optnode: Node relaxation solution
  • intsol, slpintsol: New incumbent solution
  • preintsol: Candidate incumbent solution
  • slpcascadeend, slpcascadestart, slpiterend, slpiterstart, slpitervar, slppreupdatelinearization: SLP iteration solution.
  The new function should be used instead of XPRSgetlpsol to retrieve a solution within a callback. This includes obtaining the new incumbent solution in the preintsol and intsol callbacks, and obtaining the node relaxation solution in the optnode, nodelpsolved and chgbranchobject callbacks. The corresponding presolved solution can be obtained using the new XPRSgetcallbackpresolvesolution, XPRSgetcallbackpresolveslacks, XPRSgetcallbackpresolveduals and XPRSgetcallbackpresolveredcosts functions.
• XPRSgetmipsol, XPRSgetnlpsol, XPRSgetmipsolval and XPRSgetlpsolval are now deprecated. Use XPRSgetsolution, XPRSgetcallbacksolution and related functions instead.
• Performance improvements:
  • Dual sparsification: New presolve technique to improve sparsity by column-based elimination.
  • New knapsack cover lifting procedure which uses super-additive lifting.
  • Perform additional coefficient tightening for models with big-M's.
  • Better handling of lower bound for integral objectives in objective propagation.
  • Improved sorting procedures throughout the solver.
  • New starting point heuristics have been added to the barrier algorithm when BARSTART = 1 is set.
• Minor changes:
  • New functions, XPRSsparsebtran and XPRSsparseftran, which allows sparsity in the input and output vectors to be handled efficiently. These provide the same transformation as the dense XPRSbtran and XPRSftran.
  • The new function XPRSiisprint can be used to print an IIS in the log.
  • Improved numerical handling of cutting planes to improve numerical stability of some challenging models.
  • Improved our compression routines, resulting in a notable speedup of save file generation by 30% on some platforms.
  • Functions XPRSrepairweightedinfeas() and XPRSrepairweightedinfeasbounds() will now check if the provided arrays are NULL and raise an error if so.

Xpress Nonlinear 44.01.01

• Knitro has been upgraded to its latest version of 14.1.
• Necessary reformulations (such as adding an equalscolumn) will now be restored at the end of the solve even if nonlinear presolve is off. This behaviour is affected by the NLPPOSTSOLVE control, which, if disabled, will keep the problem in a solving state at the end of the solve.
• Solution infeasibilities are now listed in the log at the end of a solve. They are also available through the attributes MAXABSPRIMALINFEAS, MAXRELPRIMALINFEAS and MAXMIPINFEAS.
• XPRSfixglobals can now also be called after global or postsolved MISLP solves to fix all MIP entities according to their values in the (nonlinear) solution.
• Performance improvements:
  • Improved performance of heuristics within multi-threaded MISLP.
• Minor changes:
  • Tokens with invalid indices will be rejected earlier and values for tokens that do not require a value (e.g., brackets) will be ignored.
  • NOFORMULADOMAINIV is deprecated.
  • The following functions have been deprecated for a while. They will now raise a deprecation warning when used: XPRSslpadddfs, XPRSslpaddtolsets, XPRSslpaddvars, XPRSslpchgdf, XPRSslpchgtolset, XPRSslpdeltolsets, XPRSslpdelvars, XPRSslpgetdf, XPRSslpgettolset, XPRSslploaddfs, XPRSslploadtolsets, XPRSslploadvars, XSLPadddfs, XSLPaddtolsets, XSLPaddvars, XSLPchgccoef, XSLPchgdf, XSLPchgformulastring, XSLPchgtolset, XSLPdeltolsets, XSLPdelvars, XSLPgetbanner, XSLPgetccoef, XSLPgetdf, XSLPgetformulastring, XSLPgettolset, XSLPloaddfs, XSLPloadtolsets, XSLPloadvars, XSLPremaxim, XSLPreminim
  • The nonlinear tuner will no longer try settings that keep non-convex quadratics in the XPRS problem, like bits 1 and 3 of the NLPREFORMULATE control and NLPPRESOLVE=0.
  • The TIMELIMIT control is now also forwarded to Knitro and both TIMELIMIT and XSLP_MAXTIME include both the presolve and Knitro time.
  • Improved the reading of .lp files with a large number of nonlinear expressions.
  • The maximum number of rows and columns when solving quadratic problems with the community license has been reduced from 5000 to 200.

Xpress Global 44.01.01

• XPRSaddmipsol is now supported for global solves.
• Performance improvements:
  • Improved computation of spatial branching point
  • General efficiency improvements

Xpress Interfaces 44.01.01

• Java, .NET
  • Functions XpressProblem.getSolution(), .getDuals(), .getSlacks(), .getRedCosts(), as well as Variable.getSolution(), Variable.getRedCost(), Inequality.getDual(), Inequality.getSlack() will now all raise an exception in case the queried information is not available.
  • Hashing of multi-dimensional map keys for dimensions larger than 2 now works correctly.
  • In the object based APIs some methods were shuffled and some argument types were generalized to super classes. No code changes should be required but you may have to recompile your application, especially if you used reflection on the object based API.
  • Function XPRSprob.createBranchObjectFromGlobal() was marked deprecated. It will be removed in a future release.
  • The MultipleKnapsack example was split into two examples: MultipleKnapsack_Arrays and MultipleKnapsack_Collections.
  • The PolygonModeling.cs example file was renamed to PolygonObjects.cs to match the name of the class and the corresponding example in Java.
• Python:
  • Using deprecated functionality now causes a warning to be emitted. Warnings can be disabled with warnings.simplefilter or warnings.filterwarnings; see the documentation for more information.
  • New deprecations: creating unlinked variables, sharing unlinked variables between several problems, passing variables or constraints objects to the API while in a callback or presolved, problem.getObjVal, accessing the constraint body via constraint.body.
  • New function constraint.getBody can be used to get an expression representing the left-hand side of a constraint.
  • Functions problem.getSlack, problem.getDual and problem.getRCost have been renamed to problem.getSlacks, problem.getDuals and problem.getRedCosts.
  • The problem.getSolution, problem.getSlacks, problem.getDuals and problem.getRedCosts functions now behave consistently with XPRSgetsolution, XPRSgetslacks, XPRSgetduals and XPRSgetredcosts.
  • The xpress.Dot operator now supports SciPy matrices of type int32, int64, float32 and float64 in CSR and CSC format.
• In R, the xprs_getsolution function is now deprecated. Instead, use getsolution, which can now be called with just a problem argument, along with getslacks, getduals and getredcosts.
• The Matlab interfaces for Xpress (Solver and Mosel) are now available for macOS on Apple Silicon M-series processors.
• The C and Java example programs can now be run more easily, with "make all.run".

Xpress Kalis 13.4.0

• Performance improvement to define idle time resources (function 'updateDurationWithIdleTimes').
• New parameter 'kalis_max_min_bound_consistency' to control the propagation level of maximum/minimum constraints.

• New parameter 'kalis_task_variables_domain_type' to control the domain representation of task variables.

  Using a bound domain only can improve performance on problems where constructive approaches are used (e.g. earliest start time heuristic).

• Performance improvement for collecting solution values (function 'getsol').

Xpress Mosel 6.8.0

Language

• New annotation property 'noimplicit'.

mmsystem 2.14.1

• 'untar' and 'tarlist' now ignore extended tar flag 'g' instead of failing.

advmod

• The package advmod is now deprecated and will be removed in a future release. Please see section Logical constructs of the MIP formulations whitepaper on how to rewrite logical expressions with indicator constraints.

deploy 3.2.1

• Executables produced from a model not using 'deploy' may now have their parameters set via the command line.

hadoop 0.0.3 (Linux platforms only)

• Updated to Hadoop 3.
• This module is no longer included in release builds. It remains available from the Mosel open source repository.

moseltest 0.10.1

• Added parallel execution of tests:
  • New parameters NBW, ASYNCOUT, LOGWKR, XSRVLOG.
  • New tags 'after', 'sync'.
• Added selection of tests by labels:
  • New parameters SKIPLABELS, ONLYLABELS.
  • New tag 'labels'.
• Added optional coloured display with default report format (new parameter COLOR).
• Coverage report for Jenkins in Cobertura format (enabled with bit 2/value 4 for COVREP).
• This tool is now included with the Xpress binaries in the release build. It also remains available from the Mosel open source repository.

Xpress Workbench 3.14

• macOS: Workbench now ships in both the Intel and ARM editions of the Xpress installer
  • Workbench binary components, unlike the Xpress libraries, are still built to target Intel Macs, and run on ARM Macs via Apple's Rosetta2 emulation.
• Allow arguments for custom Python environment in Compile Options.
  • Also allows for the executable Path to be to Conda instead, and the arguments to specify a Conda environment.
  • Can also use "$CONDA_EXE" (without quotes) to use CONDA_EXE environment variable (if present). as the executable path instead of hard-coding a particular path.
• Support for Insight 4 is removed.
  • This also means that logging in to an Insight server is now done via the OS credentials manager, rather than via username & password.
• DMP: switched from miniconda to micromamba; should not affect experience of configuring an environment with the environment.yaml file, but the resulting console output may differ from previous version.
  • Known issue: the current micromamba release may issue a spurious warning about a missing "root prefix" option, and the defaulting to the user's home directory; this can be ignored.
• Insight pane has new Manage Connections dialog for more information about Insight connection (replaces the Unlink menu item).
• Profiler now supports Python as well as Mosel models.
  • Uses the yappi Python library.
  • Unlike Mosel profiler (which puts out a .prof file for each source file), generates a single .prof file with links to the given source file.
• Updated View Designer to support latest features of VDL 5 in Xpress Insight 5.12.
• Added the "Shelf modify" action to View Designer
• Updated VDL templates to specify VDL version 5.6

Fixed in Release 9.5

Xpress Optimizer 44.01.01

• Fixed handling of infeasible start solutions in LNS heuristic.
• Fixed numerical issue in MIR tightening.
• Fixed an issue where incorrect rows were reported as a reason for infeasibility during presolve.
• Fixed an issue with wrongly calculated solution infeasibilities when solving with PRESOLVE=0
• All fixes from Xpress 9.4.3 (Optimizer 43.01.04).

Xpress Nonlinear 44.01.01

• Fixed an issue where the SLP optnode callback was never thrown.
• Fixed an issue where single-threaded MISLP solves were not postsolved correctly.
• Fixed a performance issue in the nonlinear data structures that could impact large nonlinear problems
• All fixes from Xpress 9.4.3 (Optimizer 43.01.04).

Xpress Interfaces 44.01.01

• All fixes from Xpress 9.4.3 (Optimizer 43.01.04).

Xpress Global 44.01.01

• All fixes from Xpress 9.4.3 (Optimizer 43.01.04).

Xpress BCL 44.01.01

• All fixes from Xpress 9.4.3 (Optimizer 43.01.04).

Xpress Mosel 6.8.0

mmxnlp 34.5.3

• 'getparam' now works with Knitro parameters, such as 'xktr_param_opttol'.
• Knitro parameters are now also available via the 'xprs_knitro_param' prefixes.
• Fixed an error that occurred when a multi-objective problem has 'nlctr' objectives but they are all linear.

Fixed in Release 9.4.3

Xpress Optimizer 43.01.04

• Fixed an issue that could result in MIP callbacks being fired from a different thread than the main thread when setting CALLBACKFROMMASTERTHREAD=1 and THREADS=1.
• Fixed an issue with presolving of MIPs with indicator constraints that could result in invalid tightenings.
• Fixed an issue when setting OBJSCALEFACTOR for MIQPs that could result in a sub-optimal solution being returned.
• In multi-objective problems, the beforeobjective and afterobjective callbacks are now fired for empty objectives with non-zero weight. Objectives with zero weight are considered disabled and will continue to be skipped.

Xpress Nonlinear 43.01.04

• The TIMELIMIT control is now also forwarded to Knitro and both TIMELIMIT and XSLP_MAXTIME include both the presolve and Knitro time.
• Fixed a potential crash on nonlinear problems with binary rows with only a few repeating coefficients (where the configuration presolver would have been applied).
• Fixed an issue with reading in .lp files with nonlinear coefficients like y * { x ^ 2 } <= 1 .
• Fixed an issue that could lead to a wrong value for the objective transfer column when the objective coefficient was not +1.
• Fixed an issue with the TIME attribute not including all of nonlinear presolve.

Xpress Interfaces 43.01.04 (Python, R, Matlab, .NET, Java)

• Fixed an issue that caused API calls that change nonlinear formulas to be recorded incorrectly when API recording is enabled.
• In Java and C# object based API any object information is now correctly dropped if the problem is reloaded via readProb() or loadLp() and friends. Previously, that information was not dropped which could lead to stale information being present.

Fixed in Release 9.4.2

Xpress Optimizer 43.01.03

• Fixed an issue where a wrong solution could be returned by XPRSgetlpsol in a preintsol callback fired before the initial LP relaxation has been solved.
• Improved the numerical handling when presolving MIPs with very poorly scaled coefficients.
• Fixed a deadlock that could occur when querying attribute values from callbacks.
• Fixed an issue in the barrier algorithm which could cause a crash.
• Fixed an issue where the LP relaxation solution was not available after a MIP solve that was completely presolved away.
• Fixed an issue in the mod2 presolver.
• Fixed a numerics issue in a presolver that can lead to wrong results.
• Improved numerical handling with large bound values.
• Fixed issue with restarts in component presolver.
• Improved error message when a duplicate name error occurs.

Xpress Nonlinear 43.01.03

• Fixed an issue on problems with non-convex quadratic objective and pre-existing objective transfer row that could lead to wrong results.

Xpress Global 43.01.03

• Fixed a potential crash on problems with complementarity constraints.
• Fixed an incorrect NLP-status for global solves that were stopped prematurely.

Xpress Interfaces 43.01.03 (Python, R, Matlab, .NET, Java)

• Fixed a crash when calling XPRS.license(int) from Java and .NET.
• In Python, fixed some issues with problem.getcutlist. It will now correctly return the list of loaded cuts and return the number of cuts.
• In Python, problem.getSolution now raises an error after interrupting a MIP solve before a solution is found.
• On Intel platforms, the Python interface requires OpenMP 2024.1 or earlier, due to an incompatibility in 2024.2.
• Fixed a performance regression in vectorised operations, including xpress.Dot.
• Fixed handling of duplicate terms in XpressProblem.setObjective() in case a LinTermList or a QuadTermList where passed.

Xpress Mosel 6.6.1

mmodbc 3.12.2

• 'SQLdataframe' now updates parameter 'SQLrowxfr'.

mmsheet 1.14.1

• I/O drivers csv/xls/xlsx now accept to process a 1-line range when reading with option 'skiph'.

mmxprs 4.4.2

• Constant part of objective sometimes had the wrong sign when formulated with 'mmquad'.
• Errors are no longer printed when 'writeprob' is called from a callback.

dmp 1.5.1

• Fixed an issue with the 'name' attribute being set to the wrong value by 'dmpinitcomp' in Insight 5.

Fixed in Release 9.4.1

Xpress Optimizer 43.01.02

• With numerical emphasis activated (i.e. with the NUMERICALEMPHASIS control set to any value greater than 0), a more careful numerical handling of cutting planes is applied to improve the numerical stability of some pathological models.
• Fixed numerical issue in presolve reduction that can lead to incorrect results on extreme inputs
• Fixed an issue in the crossover which could cause a crash.

Xpress Nonlinear 43.01.02

• Fixed a potential issue when reading in problems with an explicit equalscolumn from .lp-files.
• Fixed an issue where validationindices in the convergence callbacks could reflect a heuristic instead of the iteration solution.

Xpress Interfaces 43.01.02 (Python, R, Matlab, .NET, Java)

• The Xpress solver libraries have been moved out of the main Python package (xpress) into a dependency package (xpresslibs). There are no changes to the installation procedure: the xpresslibs package will be downloaded and installed automatically by Pip and Conda.
• In Conda environments, the Xpress solver libraries are now installed into the shared library folder: \Library\bin on Windows, /lib on macOS and Linux.

Xpress Mosel 6.6.1

Compiler Library

• An issue has been fixed in the handling of procedure reference types coming from packages.

Xpress Workbench 3.13.1

• Fixed issue where --prod command-line option was not correctly surfaced in xpwinsightcmd.

New in Release 9.4

Release 9.4 introduces new object-oriented .NET and Java interfaces for the Xpress solver. The Python API of the solver has also seen major updates; it now supports variables that are linked to the problem.

Xpress Solver Interfaces 43.01.01 (Python, R, Matlab, .NET, Java)

Xpress Mosel 6.6.0

• New function 'iniobuffer'.
• Operators 'min' and 'max' can now be applied to native types (with dedicated support).

• New option '-f' for command 'examine' to select directly a BIM file.

• Expressions 'x^2' and 'x^3' are now rewritten as 'x*x' and 'x*x*x' (the latter are faster to execute).

• Array assignment is now possible between arrays of different internal representations (e.g. dynamic to static).
• Posix: a model waiting for input from the console can now be interrupted.

• New function 'getstreambuffer'.
• New native type functions @2 and @3 to enable support for min/max operators.

• mempipe driver: new option 'nonblock'.

• New option 'dataframe' for xls, xlsx and csv drivers.

• 'tar' format now supports files larger than 8Gb.
• New function 'getchar(text)' (equivalent to 'getchar(text,1)').
• The types 'time', 'date', 'datetime' and 'text' now support 'min' and 'max' operators.

• New function 'getname(logctr)'
• New function 'getindictype(logctr)'

• Added Pandas 2 support.
• Added Python 3.12 support.
• Removed Python 3.5 and 3.6 support.

Xpress Workbench 3.13

• View Designer:
  • Updated View Designer to support VDL 5.4, new views will specify VDL 5.4 as the required minimum version.
  • View Designer: Added the VDL Action, FICO Drive Open.
• DMP: can filter list of Xpress Insight servers to those in the current solution.
• xpwinsightcmd: added --prod option for publishing projects to production.
• More friendly reporting of non-OK HTTP responses on first contact with Insight server.
• New Compile Option dialog entry allowing configuration of Python as a per-project setting.

Fixed in Release 9.4

Xpress Mosel 6.6.0

mosel

• Debugger: displaying a loop index over a constant set no longer produces error messages.

Compiler Library

• An issue has been fixed in the processing of private record types of imported packages that could cause errors when loading the model.

Runtime Library

• A potential deadlock has been fixed in the handling of remote files.

mmsheet 1.14.0

• xls/xlsx drivers: the default range (for "[]") no longer includes an additional row and column.

mmxprs 4.4.0

• After calling 'fixmipentities' the problem is no longer handled as a MIP.

New in Release 9.3

Xpress is now supported on Redhat Enterprise Linux 9.

Xpress Mosel 6.4.4

Fixed in Release 9.3

Xpress Mosel 6.4.4

New in Release 9.2.5

Xpress Solver and Xpress Mosel are now available for Apply Silicon M-series processors, on MacOS 11.0 (Big Sur) or later.

Fixed in Release 9.2.5

Xpress Mosel 6.4.3

Fixed in Release 9.2.4

Xpress Interfaces 42.01.03 (Python, R, Matlab, .NET, Java)

• Functions to query solution information (value, slack, reduced cost, dual) for single items were fixed in the Java and C# APIs.
• Fixed a potential race condition in BCL cleanup in .NET.
• Fixed a potential double-deletion/crash in cleanup in BCL Java.
• Fixed a Python error when adding a large number of variables to a problem.
• Python functions now raise a MemoryError when they fail to allocate memory.
• Python 3.7 is no longer supported since it has reached end of life.

New in Release 9.2.3

This release is incremental to Release 9.2.2, it does not include updates from 9.2.1.

Xpress Mosel 6.4.2

Fixed in Release 9.2.3

Xpress Nonlinear 42.01.02

• Fixed an issue with nonlinear indicators with exactly one linear nonzero.

Xpress Global 42.01.02

• Fixed an issue in presolve for problems with both convex quadratics/SOCPs and non-convex quadratics or general nonlinearities.

Xpress Mosel 6.4.2

Fixed in Release 9.2.2

This release is incremental to Release 9.2.0, it does not include updates from 9.2.1.

Xpress Mosel 6.4.1

New in Release 9.2.1

Licensing (xprl) 3.3.0

• License server connections are encrypted with TLS, if OpenSSL libraries are available.

New in Release 9.2

Major Change for Nonlinear Problems

Xpress Global is now the default solver (if licensed) for nonlinear problems. As a result, non-convex problems will be solved to global optimality. However, if a (faster) local solve is desired or if dual values are needed for continuous nonlinear problems, you can set the control XPRS_NLPSOLVER to 1 to get the problem solved by Xpress Nonlinear (if licensed). For more details, please refer to the Xpress Global Userguide.

Xpress Optimizer 42.01.01

Xpress Global 42.01.01

Xpress Nonlinear 42.01.01

Xpress Solver Interfaces 42.01.01 (Python, R, Matlab, .NET, Java)

Xpress Mosel 6.4.0

Language

mosel

xprmsrv

Compiler Library

Runtime Library

Native Interface

mmjobs 2.22.4

mmnl 2.6.1

mmodbc 3.12.0

mmreflect 1.4.0

mmssl 1.8.0

mmxnlp 34.5.0

mmxprs 4.2.0

dmp 1.5.0

executor 1.2.0

mosjvm 1.2.4

s3 0.2.0

math 1.0.0

Xpress Workbench 3.10

Fixed in Release 9.2

Xpress Optimizer 42.01.01

Xpress Global 42.01.01

Xpress Nonlinear 42.01.01

Xpress Solver Interfaces 42.01.01 (Python, R, Matlab, .NET, Java)

Xpress BCL 42.01.01

Xpress Mosel 6.4.0

mosel

Compiler Library

mmodbc 3.12.0

dmp 1.5.0

Xpress Workbench 3.10

Fixed in Release 9.1.2

Xpress Mosel 6.2.2

dmp 1.4.0

Fixed in Release 9.1.1

Xpress Optimizer 41.01.08

Xpress Nonlinear 41.01.08

Xpress Solver Interfaces 41.01.08 (Python, R, Matlab, .NET, Java)

Xpress Mosel 6.2.2

dmp 1.4.1

python3 2.1.7

New in Release 9.1

Xpress Kalis 13.3.0

Xpress Mosel 6.2.2

Fixed in Release 9.1

Xpress Mosel 6.2.2

Known issues in Release 9.1

New in Release 9.0.2

Xpress Mosel 6.2.1

Fixed in Release 9.0.2

Xpress Mosel 6.2.1

Fixed in Release 9.0.1

New in Release 9.0

Xpress Optimizer 41.01.01

• MIP Performance improvements:
  • Dynamic adjustment of the reliability threshold for pseudo cost updates. By default, strong branching is performed only on infeasible MIP entities whose pseudo costs is considered unreliable. The threshold used to determine if a pseudo cost is reliable is now dynamically adjusted during the solve, to favor a better control of the computational effort spent in strong branching.
  • Improved branching strategies on MIP models with highly degenerate objective function.
  • Separation of Strong Chvatal-Gomory cuts. The separation of Strong Chvatal-Gomory cutting planes has been introduced. The separation of these cuts can be disabled through the controls CUTSELECT and TREECUTSELECT.
  • Separation of Farkas cuts. The separation of cutting planes from Farkas infeasibility proofs has been introduced. The separation of these cuts can be disabled through the controls CUTSELECT and TREECUTSELECT.
  • A new heuristic was introduced to run at the root node parallel to the cut loop. Solutions found by this heuristic are marked by the letter 'P' in the log. This heuristic is called "feasibility jump" and can be controlled with the FEASIBILITYJUMP control (0=off, 1=on (default)).
  • The new configuration presolver attempts to reformulate binary rows with very few coefficients. The presolver drops each applicable original row in favor of a tighter formulation with new binary variables and rows modeling the choice between extremal feasible configurations of the original variables. This presolver is enabled by default. It can be disabled setting the new control PRECONFIGURATION=0.
• Added support for multi-objective optimization:
  • It is now possible to specify multiple objectives for any problem type supported by the Optimizer, such as linear, quadratic, general nonlinear optimization, and their respective mixed-integer versions.
  • Objectives can be blended using linear weights, solved in lexicographical order by specifying priorities, or any combination of the two.
  • For each objective, a user can specify various parameters when fixing the objective for lexicographical optimization: a lexicographical priority, blending weights, and an absolute and relative tolerance.
  • All solver attributes can be queried for each objective solved.
  • For further detail, refer to section 5.12 Multi-Objective Optimization of the FICO Xpress Optimizer Reference Manual.
• User-provided MIP solutions are now accepted immediately after presolve. The usersolnotify callback will be invoked with the NODEDEPTH attribute set to 0 (previously it was 1), and the CURRENTNODE attribute set to 1.
• Extended the effect of NUMERICALEMPHASIS to some bound tightening reductions. This can help to avoid wrong answers in case of numerically challenging problems.
• A new control CHECKINPUTDATA makes Xpress check input data for invalid values when supplied through the API. This checks for example for NANs in floating point arrays. It also checks for INF values in arrays where those values make no sense. The default value for the control is 1 (on). Set to 0 to disable the checks.
• A new callback has been introduced for MIPs, set by XPRSaddcbnodelpsolved. It is fired after the LP relaxation of a branch-and-bound node is solved for the first time, including the initial root LP relaxation.
• Modifying a node problem during a MIP solution in any other callback than prenode, nodelpsolved, cutmgr or optnode will result in an error.
• The MAXCUTTIME, MAXSTALLTIME and TUNERMAXTIME controls now have type double (was integer). They can still be accessed with routines that access integer controls. Values will be rounded/truncated appropriately in this case.
• The TIME and MIPSOLTIME attributes now have type double (was integer). They can still be queried with routines that access integer attributes. Values will be rounded/truncated appropriately in this case.
• New double controls TIMELIMIT and SOLTIMELIMIT. TIMELIMIT is an absolute limit on execution time (akin to negative values of MAXTIME), whereas SOLTIMELIMIT is a soft limit that only applies when a MIP solution has been found (akin to positive values of MAXTIME).
• New double control REPAIRINFEASTIMELIMIT. Imposes a time limit for the repairinfeas procedure and will replace REPAIRINFEASMAXTIME.
• MAXTIME and REPAIRINFEASMAXTIME are deprecated.
• Removed deprecated controls: CSTYLE, EXTRAPRESOLVE, EXTRAQCELEMENTS, EXTRAQCROWS, GLOBALFILEBIAS, HEURSTRATEGY, LINELENGTH, LPTHREADS, PERTURB, TUNERROOTALG.
• Renamed controls and attributes:
  • GLOBALFILE[SIZE,USAGE,LOGINTERVAL] is now TREEFILE[SIZE,USAGE,LOGINTERVAL].
  • MAXGLOBALFILESIZE is now MAXTREEFILESIZE.
• Renamed API functions:
  • XPRSfixglobals() is now XPRSfixmipentities().
  • XPRSgetglobal() is now XPRSgetmipentities().
  • XPRSloadqcqpglobal() is now XPRSloadmiqcqp().
  • XPRSloadqglobal() is now XPRSloadmiqp().
  • XPRSloadglobal() is now XPRSloadmip().
  • XPRS[add,remove]cbgloballog() is now XPRS[add,remove]cbmiplog().
• Removed deprecated API functions: XPRSglobal() and XPRSinitglobal()
• XPRSgetsol() has been deprecated.

Xpress Mosel 6.2

Language

• The effect of procedure 'setname' now applies independently of compiler options.
• New version of 'setname' for arrays.
• New one-line syntax for 'if' statement: if condition: statement
• Arrays indexed by constant texts can now be directly indexed by strings (e.g. A('a') can be used in addition to A(text('a')) ).
• The clone operator can now be applied to array/list/set types supporting assignment.

Mosel

• New compiler option '-wu' to report public unused symbols (in addition to private unused symbols).
• Profiler: more than 127 source files can now be handled during a profiling run.

Compiler

• The compiler now reports private unused symbols as warnings.
• Documentation generation now supports unions and subroutine references.
• Type description in error messages has been improved.

Runtime Library

• The logic of automatic unloading of modules has been revised in order to avoid situations where a module is unloaded and then reloaded immediately.
• Windows support for long path names.

Native Interface

• New function 'gettypeid' to retrieve the type ID of a set/list/array/union.
• New function 'hmenum' to enumerate elements of a hashmap.
• New function 'setentname' to associate a name with an entity.
• New type property XPRM_DTYP_NAMED to declare a native type supporting name association (and dedicated behaviour of the 'create' function).

deploy 3.0.0

• The module can now access the arguments of the 'mosel' command following the marker '--' (e.g. mosel mymodel -- arg1 arg2)
• New procedure 'shiftargv'.
• New I/O driver 'skiphead'.

mmhttp 3.6.1

• HTTP server: response headers now use CRLF as line termination.
• Windows support for long path names.

mmjobs 2.22.3

• getstatus(Mosel) now returns -3 if the instance reported an error while disconnecting.
• Windows support for long path names.

mmnl 2.6.0

• API update for Mosel 6.2.

mmreflect 1.2.0

• New subroutines 'getallfields', 'getfieldname', 'getfieldtype', 'findfield', 'getfieldval', 'setfieldval' for accessing record fields.

mmsheet 1.12.0

• New options for I/O drivers xls/xlsx: 'skiperrcell;inf=;minf=;nan='

mmsystem 2.12.2

• Improved handling of UTF8 characters in regular expressions ('.' can now match a non-ascii symbol).
• Windows support for long path names.

mmxprs 4.0.0

• New function 'getmatstat'.
• New control parameter 'xprs_usechecktime' (to use the callback checktime to interrupt a solving procedure).
• Support for optimisation with multiple objectives:
  • new versions of: minimise/minimize, maximise/maximize, loadprob
  • new routines: 'getobjintattr', 'getobjrealattr'
  • new type: 'objconfig'
  • new routines for accessing 'objconfig': get|setpriority, get|setweight, get|getabstol, get|setreltol

dmp 1.4.0

• New overload of dmpiniturl procedure will take a function pointer that returns authorization headers.

python3 2.1.5

• Added Python 3.10 support.

Xpress Workbench 3.7

• View Designer
  • Upgraded to VDL version 5.0 for all new VDL files created from a template or the quick start app.
  • Support for VDL 4.8 files remains for existing VDL views.
  • Added Tableau component, which enables you to configure one or more Tableau visualizations within your view.
  • VDL actions for data manipulation is available from the Palette.
• Tableau Workbooks
  • Tableau Workbooks: once published to an Insight 5 server, Tableau Workbooks are no longer synched between Workbench, Insight, and Tableau. A Full Publish to an existing Insight 5 app will now remove any Workbooks that are no longer present in Workbench.
• Insight Companion Files
  • Insight Companion files created from File menu or via the Views control on the Insight pane will now be written for Insight 5 apps. There are comments in the newly-created Companion file, advising users of earlier versions of Insight of the changes they will need to make in order to downgrade.
• When launching a Mosel process, PATH and related environment variables have the relevant location for the Xpress installation applied to the start, so they take precedence over any others that may be in the environment variable already. (Previously, the Xpress installation folders would not be added to the start of the path if they were present already.)
• Updated several 3rd-party dependencies.

Fixed in Release 9.0

Xpress Optimizer 41.01.01

• Fixed a problem in which the multi-threaded RINS heuristic at the root node may have used more threads than intended.
• All fixes from Xpress 8.14.3 (Optimizer 40.01.03).

Xpress Nonlinear 41.01.01

• All fixes from Xpress 8.14.3 (Nonlinear 40.01.03).

Xpress Solver Interfaces 41.01.01 (Python, MathLab, .NET, and Java)

• All fixes from Xpress 8.14.3 (Interfaces 40.01.03).

Xpress BCL 41.01.01

• All fixes from Xpress 8.14.3 (BCL 40.01.03).

Xpress Kalis 13.2.3

• Solution query when using the two phases scheduling procedure no longer crashes.

Xpress Mosel 6.2

Compiler Library

• An issue has been fixed in the handling of array dereferencing using sets of constants.
• Importation of packages defining subroutine references on local subroutines no longer produces invalid code.

Runtime Library

• An issue has been fixed in the handling of locally defined decision variables.
• A potential failure when exiting the program has been fixed.

Remote Invocation

• A potential memory corruption has been fixed in the conversion of very large real numbers to their textual representation.

Xpress Workbench 3.7

• (Desktop, macOS) The long path name issue referred to in the 3.6.2 section of these notes has been ameliorated; it now only affects *extremely* long paths that will not fit on the command-lines of some executables.
• Fixed: (Desktop) Issue when opening multiple projects at once would fail due to network ports not being available.

New in Release 8.14.2

Xpress Solver 8.14.2 (40.01.03)

Xpress Optimizer

• Fixed an issue in presolve that could introduce small constraint violations.
• Fixed an issue in presolve for mixed-integer quadratic problems.
• Fixed an issue where some messages from the nonlinear solver were not sent to the optimizer message callback.
• Fixed an issue in the console optimizer where upper case commands incorrectly processed escape characters in their arguments.

Xpress Nonlinear

• Fixed an issue where terms could incorrectly be dropped when nonlinear presolve was disabled.
• XSLPcopyprob no longer copies controls or callbacks, as documented.

  To copy controls and callbacks, use XSLPcopycontrols and XSLPcopycallbacks, respectively, after copying the problem.

  Note that XSLPcopycontrols will not copy convergence criteria or Knitro controls. These should be set manually.

Xpress Interfaces (Python, R, Matlab, .NET, Java)

• [Python] The Nonlinear callback function 'preupdatelinearization' is now available through Python.
• [Python] Nonlinear user functions can now return derivatives on demand. The optional argument 'derivatives' to xpress.user() can now take one of the values: "never", "always" or "ondemand".
• [Python] Fixed an issue where problems created using problem.copy() could crash when solved, due to callbacks being copied incorrectly.
• [Python] Performance improvements which reduce the time taken to create large numbers of variables.

New in Release 8.14.1

Xpress Solver 8.14.1 (40.01.02)

New in Release 8.14

Xpress Solver Interfaces 40.01.01 (Python, R, Matlab, .NET, Java)

Xpress BCL 4.14.0

Xpress Kalis 13.2.0

Xpress Mosel 6.0

Runtime Library

• New type properties 'TPROP_NBELT' and 'TPROP_SIGN' (cf 'gettypeprop').
• New environment variable 'MOSEL_TMP' to specify the temporary directory for Mosel.

Fixed in Release 8.14

Xpress Optimizer 40.01.01

Xpress Mosel 6.0

Fixed in Release 8.13.5

Fixed in Release 8.13.4

New in Release 8.13.3

Xpress Mosel 5.8.2

executor 1.1.1

mosjvm 1.2.2

s3 0.1.2

Fixed in Release 8.13.3

Xpress Mosel 5.8.2

New in Release 8.13.2

New *BETA* AArch64 port of Xpress Optimizer , Xpress Nonlinear and Xpress Mosel, supporting ARMv8-A on Linux (such as Graviton2 instances on Amazon Web Services)

Xpress Optimizer 39.01.03

New in Release 8.13.1

Xpress Optimizer 39.01.02

Xpress Nonlinear 39.01.02

Xpress Solver Interfaces 39.01.02 (Python, R, Matlab, .NET, Java)

Xpress Mosel 5.8.1 Runtime Library

mmhttp 3.4.1

mmjobs 2.22.1

mmreflect 0.0.3

mmxml 2.2.9

Xpress Workbench 3.5.1

Trimmed excessively verbose logging

New in Release 8.13

Fixed in Release 8.13

Fixed in Release 8.12.4

Xpress Optimizer 38.01.05

Xpress Solver Interfaces 38.01.05 (Python, R, Matlab, .NET, Java)

Fixed in Release 8.12.3

New in Release 8.12.2

Fixed in Release 8.12.2

New in Release 8.12.1

Fixed in Release 8.12.1

New in Release 8.12

Fixed in Release 8.12

Back to top

New in Release 8.11.1

Xpress Mosel moduless3 0.1.0

dmp 1.1.1

Fixed in Release 8.11.1

• Fixed an issue related to retrieving solutions to non-convex quadratic problems after nonlinear post-solve.

New in Release 8.11

Xpress Mosel 5.4.0

Fixed in Release 8.11

Mosel 5.4.0

Fixed in Release 8.10.1

Xpress Mosel 5.2.4

Runtime Library

New in Release 8.10

Xpress Mosel 5.2.2

mmnl 2.2.0

Fixed in Release 8.10

Xpress Mosel 5.2.2

Fixed in Release 8.9.4

Xpress Mosel 5.2.3

Fixed in Release 8.9.3

Xpress Mosel 5.2.1

Fixed in Release 8.9.2

Xpress Mosel 5.2.1

New in Release 8.9.1

New in Release 8.9

Fixed in Release 8.8.4

Xpress Optimizer 35.01.05

• XPRSalter will now change the neutral row of the objective as well if the name of the row rather then **OBJ is specified.
• XPRSalter will now read floating point numbers in full accuracy.
• Improved numerical behavior for handling free variables with the barrier algorithm.
• Fixed an issue in which a user function was not provided the context pointer.
• Fixed an issue that could cause IIS to crash.
• Fixed an issue with the analytic center presolver for MIPs, which could result in data corruption, when Barrier is used in the tree and fails to solve a node relaxation.
• Fixed an issue with presolve implications when represolving a MIP, which could result in invalid solutions being returned.
• Fixed an issue in the storage of branching history information that could affect performance.
• Fixed an issue in conflict analysis that could potentially cause too short conflicts.

• New XSLP_NLPSTATUS for capturing solve failiures, XSLP_NLPSTATUS_UNSOLVED.

Fixed in Release 8.8.3

Xpress Optimizer 35.01.04

• Added support for Python 3.8 in the Xpress Python interface; support for Python 3.4 was dropped. Python 2.7 is still supported.
• A new control, CROSSOVERFEASWEIGHT, has been added to change crossover behaviour.
• Fixed and issue with handling of indicator constraints when presolve is turned off that could result in violated solutions being returned.
• Fixed an issue in implied entity detection if dual reductions are disabled
• Improved the deterministic timer for primal simplex.
• Fixed a performance issue with the diving heuristic for large MIPs with few non-binary variables.
• Fixed in issue in IIS that could cause a problem for infeasible MIP instances with range rows.
• Fixed an issue involving QCQP problems that are convertible to QP instances.
• Improved numerical behavior for estimaterowdualranges.
• Fixed an issue that could lead to SOS-constraint with non-positive variables being dropped in presolve.

• Fixed an issue that could cause a crash when using XSLPchgccoef with longer than 500 token (string can be arbitrary long) formulas.
• Improved performance of postsolving dual solutions.

• Fixed possible memory leak when loading problems with both quadratic constraints and global entities or SOSs
• Fixed possibly wrong memory allocation size for loading names into the Optimizer when loading problems with more than 16M constraints or variables

Fixed in Release 8.8.2

• Fixed an issue that might cause reading incorrect memory areas when accessing a constraint in the Python interface.
• Fixed an issue that could cause crash when the barrier is executed with warm-start.
• Fixed an issue in presolve that could lead to returning infeasible solutions due to dropping wrong rows.

• Various minor fixes.

• Windows: a possible memory corruption has been fixed in the handling of the 'mempipe' driver.

• Invalid assignments now report a runtime error.

• 'newtar' no longer generates an invalid archive with some large directory names.

• Invalid assignments now report a runtime error.

• Fixed: Mosel processes not being cleaned up after running

Fixed in Release 8.8.1

Xpress Optimizer 35.01.02

• Fixed an issue in the Python interface whereby integer variables are converted to binary variables.
• A new control, BARFAILITERLIMIT, has been added to limit the unsuccessful barrier iterations.
• Fixed a memory leak in the loadproblem() method of the Python interface.
• Fixed an issue in the dual simplex that could cause cycling in Phase 1.
• Range row support added to XPRSalter.
• Fixed a potential memory leak on Windows.
• Fixed an issue in presolving that could lead to a wrong sign for added inequalities.
• Fixed an issue with advanced presolving of MIPs that could result in a memory corruption.

• When sending a query use the 'origin-form' for the query line (some buggy servers do not support 'absolute-form').

• A submodel can no longer be marked as not running while the corresponding end-event is not yet available.

• Support for memoryuse.

New in Release 8.8

• The Solution Refiner now includes an additional option to apply Iterative Refinement. Iterative Refinement involves shifting and rescaling of the problem between iterative solves of an LP, in order to improve a linear solution. It is based on the work described in "Iterative Refinement for Linear Programming, A.M. Gleixner, D.E. Steffy, K. Wolter, ZIB Report 15-15 (May 2015)". New bits have been introduced for the REFINEOPS control for Iterative Refinement:
  • bit 5: Enable iterative refinement
  • bit 6: Use quad precision when updating the solution.
• Memory reporting attributes have changed:
  • CURRENTMEMORY: Now reports on memory used by all objects part of the current problem solve, instead of just the queried object.
  • TOTALMEMORY: New control that reports on the total memory used by all solver objects. This includes Nonlinear and BCL.
• It is now possible to impose memory limits on a solve. New controls have been introduced:
  • MAXMEMORYSOFT: Adjust solver behavior to stay within the given limit (not guaranteed)
  • MAXMEMORYHARD: Terminate a solve if this limit is breached. The solver will return in an interrupted state that can be resumed and where solution information is available.
  • RESOURCESTRATEGY: A new bit-vector control. If bit 1 is set, then the solver is allowed to make a non-deterministic choice based on MAXMEMORYSOFT.
• A new control, MIPDUALREDUCTIONS, has been introduced, which can be used for disabling all dual related reductions throughout a MIP solve. This is particularly useful for users who do their own constraint generation or branching inside callbacks. This avoids having to turn off individual reductions that assume all constraints and the objective function are described completely as given in the initial formulation.
• It is now possible to work directly with individual bits of bit-vector controls in the Tuner. When listing bit-vector controls in a method file, use ":" instead of "=" when specifying bits instead of values to use.
• Added support for extended modelling information when reading LP and MPS file formats.
  • Added support for piece-wise linear objective functions and constraints.
  • Added support for general constraints: MIN, MAX, ABS, AND, OR
  • All such extended modelling information is converted into a linear continuous or MIP equivalent formulation at read time, potentially by introducing new constraints and variables.
• Improved crossover and simplex performance for large scale LPs.
• Presolve clique lifting and clique cutting for MIPs has been rewritten to remove scalability issues.
• Fixed several MIP performance issues for very large problems.
• Minor changes:
  • Two controls have been added for ignoring detected process control group limitations (e.g. used in "Containers") in Linux: IGNORECONTAINERSPULIMIT and IGNORECONTAINERMEMORYLIMIT
  • Extended the detection of Linux control groups (v1) to recognize more system configurations.
  • All coefficients are now written to and read from LP and MPS files exactly, meaning that when a coefficient is written to an LP/MPS file and read back in again, the exact same coefficient is recovered. This obsoletes the "z" option for the MPS file format.
  • Files written to the LP file format will now include a line identifying the version of Xpress it was written with.
  • The LP file format now supports unlimited line lengths.

• The Python interface has now a much more flexible problem.getSolution() function for evaluating expressions and aggregate Python objects.
• Optimization problems can be specified in a compact way in the Python interface by inserting variables, constraints, and the objective function in the xpress.problem() call.
• A much smaller memory footprint for the Python interface's modeling feature. The amount of memory required for allocating variables and constraint is substantially reduced.

• Redesigned formula handling when using mmxnlp:
  • Decreased memory usage when building expressions using 'nlctr' objects embedded into one another.
  • Decreased memory usage when generating identical 'nlctr' expressions.
  • Decreased peak memory usage when Mosel loads a problem.
• Improved the nonlinear eliminator in the nonlinear presolver.
• Nonlinear liminations are now enabled by default.
• Support for models with more than 2^31 nonlinear tokens.
• Parallel evaluations support for expensive blackbox functions in embedded expressions:
  • Blackbox functions depending on one another can now calculate their partial derivatives in parallel.
  • New callback 'PREUPDATELINEARIZATION'.
• New nonlinear primal integral support:
  • New attribute XSLP_PRIMALINTEGRAL' and 'XSLP_PRIMALINTEGRALFACTOR'.
• Nonlinear license will now be checked from Mosel only at the time of loading or solving a problem.

• Improved handling of quadratic expressions.
• Support for += and -= (that preserve constraint direction).

• In debug mode a warning is now displayed when strings are truncated.

• mmsqlite driver: new option 'timeout' to define the busy timeout.
• In debug mode a warning is now displayed when strings are truncated.

• 'formattext' is no longer limited to 64k text strings.

• Now supports Amazon Corretto JVM.

• The latest supported Python version is 3.7.5. Python 3.8.0 is not supported (https://bugs.python.org/issue37633).

• New Mosel syntax analysis to highlight errors in source code.
• The Outline pane now works on Mosel files, as well as Javascript, and allows to jump to definition.
• View Designer
  • Now highlights when the element hierarchy on the page breaks soft validation rules.
  • Attribute values are visible in the attributes pane when attribute is set to default
  • Element selection no longer writes a placeholder attribute into the VDL code.
  • Attributes are now organized into collapsible groups.
  • When selecting an element in the View Designer and then switching to the code editor, the code editor will jump to the selected element source
  • Copy/cut/paste operations via the clipboard are now supported in the View Designer
  • New project preference for selecting code editor or View Designer as default editor for VDL files

• Xpress Kalis is now available for MacOS.
• The Commmunity Licence now enables a size-limited version of Xpress Kalis.
• New procedure 'setsetuptime' to specify a setup time between two tasks iff they both require the same resource.
• New function 'dot' for efficient computation of the scalar product between an array of decision variables and an array of numbers.

Fixed in Release 8.8

Xpress Mosel 5.0.2

• Memory usage information can now be retrieved from within a callback.

• Fix for multiagg issue with median calculation on sample-weighted data.

• Fix integer overflows in cumulative resource constraint usage.
• Fix return value of 'cp_schedule' when 'kalis_max_solutions' set to 1 and solution is suboptimal.
• Various minor memory issues and bug fixes.

• The Xpress Kalis examples 'runfapp[r].mos' fail to run properly under Linux.

Fixed in Release 8.7.3

• A new control, CROSSOVERFEASWEIGHT, has been added to change crossover behaviour.
• Added support for Python 3.8 in the Xpress Python interface; support for Python 3.4 was dropped. Python 2.7 is still supported.
• Fixed an issue that could cause crash when the barrier is executed with warm-start.
• Fixed an issue with advanced presolving of MIPs that could result in a memory corruption.
• Fixed an issue in presolve that could lead to returning infeasible solutions due to dropping wrong rows.

Fixed in Release 8.7.2

• Fixed: when publishing, Workbench would not recompile the model if an included source file was modified.
• Fixed: Workbench did not start up if home directory was unavailable.
• Fixed: model schema did not appear in view designer for some models.

Fixed in Release 8.7.1

• Fixed an issue in nonlinear formula eliminations in the nonlinear presolver.
• Limited the number of automatically allowable step bound repaired infeasible iterations.

• Fix for multiagg issue with median calculation on sample-weighted data.

New in Release 8.7

Xpress Mosel 5.0.2

Xpress Workbench 3.1.0

Fixed in Release 8.7

Xpress Mosel 5.0.2

New in Release 8.6

Fixed in Release 8.6

Known Issues in Release 8.6

Fixed in Release 8.5.11

• Made bound reduction less sensitive toward extreme bounds in log() and log10() functions.
• Improved initial point heuristics with cascading.
• Fixed a potential issue when runing parallel MISLP with Mosel user functions.
• Improved initial point heuristics.
• Improved cascading logic for network structures.
• Support Mosel user functions returning partial derivatives with variable argument lists.
• Off-scale bounds due to functions such as exp() now trigger a warning rather than infeasibility.
• New public control XSLP_DELTA_INFINITY to limit placholders taking exceedingly large values.
• Fixed in issue with the multistart search when using initial values from Mosel
• Fixed an issue with eliminations on singleton rows.

New in Release 8.5.6

Xpress Mosel modules

• Allow option to run any classes in restricted mode by setting MOSJVM_ALLOW=*
• Allow jvmclasspath parameter to be set in restricted mode

• New module to allow models running in DMP to call other DMP resources in solution

• When running within DMP, authorization will no longer expire after 1 hour
• Failed HTTP requests will now be retried

Fixed in Release 8.5.6

• Fixed an issue with eliminations on singleton rows.
• Fixed a memory leak that could occur when repeatedly restoring and presolveing a problem.
• Calling XSLPglobal now resets the solve timer.
• Fix incorrect header in save files.
• Support cubic root of negative numbers.

• An issue has been fixed in the handling of the connection pool

• 'fcopy' with 4 arguments now closes all its streams after an error

• Added cancel button to progress dialog shown when publishing to Insight.
• Fixed an issue where the progress dialog sometimes got stuck when publishing to Insight.
• Fixed an issue where error messages were not displayed after disconnecting from a child instance of Mosel.
• Fixed an error, "Cannot read property 'getDocument' of undefined" when saving a tab which is not fully loaded.

• Fixed an issue that caused creation of xprl.log files on Windows

New in Release 8.5.3

• More agressive ERRORCOST updates for unbounded problems.
• New convergence option XSLP_CONVERGEBIT_NOQUADCHECK.
• Improved convergence criteria for mixed convex quadratic and nonlinear problems.

• HTTP server now rejects invalid queries

• Posix: better handling of quotes in the processing of the 'system' command

• BETA support for macOS
• Clicking a line in a Mosel stack trace opens the correct source file location.
• Subprocesses executed from Mosel inherit the correct working directory specified in Workbench.
• When autocompleting the name of a subroutine from a Mosel package or module, all available subroutine signatures are now shown.

Fixed in Release 8.5.3

• Fixed a potential issue when solving problems with integer delta variables.

• An issue has been fixed in the handling of file inclusion

• xlsx/xls/csv: an issue has been fixed in the conversion of Booleans to text

• Fixed an error when downloading very large arrays as CSV.
• Fixed an issue where Mosel error messages where not clickable when they wrapped onto a new line.
• Fixed an issue where relative source file paths passed on the Workbench command line were not correctly located.
• Fixed an issue where incorrect file tree was sometimes shown when several workspaces were open at the same time.
• Fixed an issue where 'Show item in Explorer' stopped working after compiling a model.

• Dialog appears intermittently while editing source files: "File has been changed on disk."
• Not possible to set environment variables when running Mosel models.
• Workbench does not detect changes to the environment, e.g., XPRESSDIR, after being restarted.

New in Release 8.5

• Symmetries may now be re-detected locally during a MIP tree search when it is determined to be beneficial.
• Improved branching for highly symmetric MIP problems.
• Improved zero-half cuts, in particular for binary quadratic problems.
• New multi-commodity network flow cuts. A new control, NETCUTS, has been introduced for these cuts.
• Improved dual simplex perturbation scheme for highly dual degenerate MIPs.
• Improved performance of crossover through better exploitation of vectorization and CPU caching.
• New API functions XPRSgetlpsolval, XPRSgetmipsolval and XPRSgetbasisval for retrieving the solution value or basis status for an individual variable or constraint.
• Two new controls, PRIMALPERTURB and DUALPERTURB, have been introduced to replace the single PERTURB control. Perturbation can now be controlled separately for primal simplex and dual simplex.
• A new control, PRESOLVEPASSES, has been introduced to allow repeated applications of the initial preprocessing of a problem.
• The obsolete control TEMPBOUNDS has been removed.
• All fixes from Xpress Update 8.4.7 (Optimizer 32.01.10).

• Improved MINLP rounding heuristics.
• Improved SLP-MIP-SLP heuristics in case some rows are enforced in the first SLP iteration.
• All fixes from Xpress Update 8.4.7 (Nonlinear 32.01.10).

• Function 'copyval' can now be used with STR_SET, STR_LIST and STR_ARR

• New function 'tcpping' to test a server
• Unauthorized urls in FAC are now reported as error 997 (execution is no longer interrupted)
• Support for SNI (server name identification) in https

• Excel driver: better handling of files stored on network drives

• Updated version of ca-bundle.crt
• New driver 'hmac'

• Comparison operator has been added to 'time', 'date', 'datetime' and 'text' to enable comparison of records or arrays including these types

• New procedure 'loadlpsol' to load LP solutions for barrier warm start and crossover
• New procedure 'crossoverlpsol' to cross over a loaded LP solution
• New procedure 'uselastbarsol' to retrieve a barrier solution from before crossover
• Parameter 'xprs_objsense' can now be changed

• Execute Python 3 scripts from Mosel
• I/O driver 'python' for exchanging data using 'initializations from' and 'initializations to'
• Supported on Windows, Linux, MacOS

• Now a pure Mosel package (aec2 and smew modules dependencies have been removed)
• Support for Amazon signature protocol v4

• When debugging, Mosel arrays, sets and lists can be displayed in table form, where they can be sorted/filtered.
• When debugging, Mosel arrays, sets and lists can be exported in a CSV file.
• Added a compile button to the toolbar.
• Mosel compiler options can now be specified.
• Mosel models can be run with profiling or coverage enabled.
• Improved support for Mosel DSO data types in the debugger.
• Added a context menu item in the editor to find all usages of the selected model symbol.
• After running or compiling a model, the output pane is now automatically shown, if it was hidden.
• Workbench now looks in the MOSEL_DSO environment variable when searching for Mosel package dependencies.
• Compiler uses MOSEL_DSO environment variable to find dependencies
• Workbench now integrates with the online help system

• Updated.Net Scanner class to support exponential notation

Fixed in Release 8.5.0

• An issue has been resolved in the handling of assignments of loop indices referring to external types
• 'initialisations' blocks no longer accept lists of arrays of user types (this is not supported)

• A possible memory corruption has been fixed in the handling of list concatenation when the first list is empty

• An issue has been fixed in untar/unzip related to handling of directory names
• 'setenv': removing an environment variable by giving it an empty value now always works

• Fixed an issue with models that use the I/O driver without initializing R

• Fixed an issue where opening a second Workbench could hang on the loading screen.
• Fixed an error, "Current scheduler state does not fix stack level", shown when stepping in the debugger.
• Fixed an issue where the watches tree would collapse when inspecing a large array.
• Fixed a warning which occurs when many Mosel processes are created, "Possible EventEmitter memory leak detected".
• Fixed an error, "Step operation interrupted by debugger", which could occur when stepping over slow subroutine calls.
• Fixed incorrect label, "Select Folder to Upload", when creating or opening a project.
• Fixed error when opening projects stored on a drive other than C:
• Fixed issue where Mosel output contains unexpected symbols on Windows 7
• Fixed an issue where Mosel appears to hang and display no output when running a model
• Fixed issue where output pane would sometimes remain blank when publishing
• Fixed issue where watch tree collapsed immediately after being expanded

New in Release 8.4.4

• Java: new field 'XPRD.rcmdEnv' to define environment variables in the environment of the process created via 'rcmd:'

• New function 'setdsoparam'

• New version of SQLcolumns

• 'waitexpired' now returns 'true' when 'waitfor' has been called with a 0-time limit and no valid event was found

• New routines findfiles(findfiles(integer,list of text,text,text), findfiles(integer,list of text,text), findfiles(list of text,text)

• Namespace prefixes can now be used in search paths

• New module to solve the problem with an AMPL-compatible solver

Fixed in Release 8.4.4

• Improved numerical stability for problems with a numerically difficult convex quadratic objective.
• Fixed an issue where an infeasible solution could have been returned.
• XSLP_VERSION and XSLP_MINORVERSION now returns the correct value.
• XSLP_ALGORITHM bit 6 'Reset XSLP_DELTA_Z to zero when converged and continue SLP.' now behaves as intended.
• Added support for using XPRSaddcols64 and XPRSaddrows64 together with XSLP.

• 'getnextattrdesc' no longer returns invalid type numbers
• Comparisons 'infinity=infinity' and 'infinity<>infinity' have been fixed

• An issue has been resolved in the handling of empty statements in 'if' and 'case' conditions

• 'waitfor' may no longer miss an event with the expected properties when the option 'keep' is enabled
• A potential memory corruption has been fixed in the 'send' procedure when an event is sent to a terminating model

• csv driver: reading now stops when an out of memory error is detected

• A potential memory corruption has been fixed in the 'sslrandomdata' function

• An issue has been fixed in the implmentation of "textfmt(real,int,int)"

• Fixed issue where output pane would sometimes remain blank when publishing
• Fixed issue where watch tree collapsed immediately after being expanded
• On premises, the Mosel compiler now uses the MOSEL_DSO environment variable to locate packages and modules
• On premises, Workbench now integrates the online help site rather than the CHM version of the documentation

• Fixed an issue re-establishing connection to license server after a network error.
• Fixed an issue with redundant server licensing when license server started with -d flag
• Corrected error message displayed when counted license checked out 60+ times

• Fixed bug in 'produces' constraints

Fixed in Release 8.4.2

• Improved local search heuristics for MINLP problems.
• Improved relative infeasiblity checks when XSLP_FILTER bit 1 is enabled.
• The SLP iteration solution refiner will now check for change in relative primal infeasiblity.
• Added missing Knitro max nodes control export.
• XSLP_MIPITER now applies to the SLP-MIP-SLP and MIP-in-SLP heuristics as well.
• New fallback strategy for problems instable in linear presolve due to cascading structures.

• OCIcommit/OCIrollback do not set OCIsuccess

• SQLautocommit no longer ignored

• Fixed errors when running Mosel models: "died waiting for dll loading, errno 11" and "Resource temporarily unavailable"

Fixed in Release 8.4.1

• Improved barrier fallback strategy for numerically instable solves.
• The drcoltol handler and callback are now initiated by XSLPcontruct if set up.

• An issue has been resolved in the handling of conversions in string concatenations.

• HTTP headers of responses are now case insensitive.

• New default value for https_ciphers: "TLSv1.2+HIGH:TLSv1+HIGH:@STRENGTH"

• Fixed issue where old backups were not being cleaned up properly

New in Release 8.4

• It is now possible to exploit symmetries in LP problems using a procedure called LP Folding. This can significantly reduce the size of the problem to be solved and thereby speed up the solution process.
  • LP Folding is automatically applied and can be used with any LP solving algorithm.
  • A new control, LPFOLDING, has been introduced to specify whether LP Folding should be applied.
  • When LP Folding is applied a solve will now go through three phases: solve the reduced LP using the selected algorithm, a crossover phase and a final cleanup. The exception is when barrier is applied without crossover.
• For tiny MIP instances with integer or binary variables with a wide range of coefficients, it is now possible to apply a basis reduction algorithm as part of presolve, which will reformulate the integer columns using a reduced basis that is closer to ortho-normal. A new control, PREBASISRED, has been introduced to specify whether MIP basis reduction should be applied.
• More uses for the Barrier algorithm have been introduced as part of a MIP solution process. It can now be used for tightening or reducing the problems. A new control, PREANALYTICCENTER, has been introduce to specify when to use a Barrier solution.
• Several refinements to the branch-and-bound code to improve MIP performance.
• Output to the file created by XPRSsetlogfile is now flushed after each line.
• A new control, HEURFORCESPECIALOBJ, has been introduce to force alternative objective local search heuristics to run on a MIP, even when the automatic strategy has determined they should not be run.
• More choices for TUNERTARGET when tuning a MIP.
• New function, XPRSfeaturequery, to query which features are allowed by the current license.
• Quadratic constraints in binary variables can now be linearized when this appears beneficial.
• All Xpress-Nonlinear functionality is now included in the XPRS library file. It is no longer necessary to link against the XSLP library.

• Long file names are no longer truncated when reporting error messages

• LP/MIP matrix manipulation routines are now avaiable from the NI
• New service 'PROVIDER'

• New tag 'recfldval' for documeting record field values

• New functions 'erf' and 'erfc'

• New procedure 'hasfeature' to check license features

• New versions of 'svgwaitclose' to change behavior on browser window closing in Workbench
• New style property 'SVG_ANIMATE'

• Compiler errors are highlighted and can be clicked to jump to the source location
• Compiler looks for bim file dependencies in the same directory as the source files
• If a model raises an error while debugging, the debugger stops at the line where the error occurred
• More Mosel module symbols are autocompleted
• Insight sidebar is automatically opened for Insight projects only
• Improved error reporting when Mosel model crashes
• Improved size and position of Workbench window on low resolution displays
• Improved error reporting when Mosel graph cannot be reopened

• Significantly improved performance on numerically challenging problems.
• Added support for the min/max functions both in Mosel and through API.
• Added support for the erf (standard error function) and its complement erfc to the API and Mosel.
• Work with best solutions: XSLP_FILTER bit 1 is now default. Checks now include the initial solution.
• New attribute XSLP_SOLSTATUS to indicate if a solution is available. This attribute complements XSLP_NLPSTATUS.
• New attribute XSLP_VALIDATIONSTATUS to check if the current solution is deemed feasible, set up by XSLPvalidate.
• New attribute XSLP_TOTALEVALUATIONERRORS to indicate the number of evaluation errors that occurred during a solve.
• New API function XSLPnlpoptimize to solve the problem using the existing optimization sense.
• New option for XSLP_HEURSTRATEGY to include all heuristics with no effort limits imposed.
• Mosel can now read and write solutions of nonlinear problems using 'writesol' and 'readsol'.
• All Xpress-Nonlinear functionality is now included in the XPRS library file. It is no longer necessary to link against the XSLP library.
• Deprecated functions in this release:

  XSLPloadivfs, XSLPaddivfs, XSLPchgivf, XSLPdelivf, XSLPgetivformula,
  XSLPloadxvs , XSLPaddxvs, XSLPchgxv, XSLPchgxvitem, XSLPdelxvs, XSLPgetxv,XSLPgetxvitemformula,
  XSLPloaddcs, XSLPadddcs, XSLPchgdc, XSLPdeldcs, XSLPgetdcformula, XSLPopt,
  XSLPsetcbmessageF, XSLPsetcbitervarF, XSLPsetcbcascadevarF

• Implemented Mosel restrictions

Fixed in Release 8.4.0

Xpress Mosel 4.8.0

• An issue has been fixed in the handling of extended types in packages

• Java: an issue has been fixed in the handling of connection string parameters

• Objective modification now always reported to optimiser

• An issue has been fixed in the 'hex' driver when used for reading

• JSON reader: possible failure when loading a file containing only an empty string

• Using mmquad no longer prevents optimising linear objectives

• Fixed issue where arrays of arrays could not be inspected in the debugger
• Fixed issue where cursor was sometimes drawn in the wrong place when editing source files
• Fixed issue where model output was truncated after 5000 lines
• Fixed issue where variables tree could not be resized
• Fixed issue where Mosel annotations were only autocompleted once
• Fixed long delays when saving files, running Mosel models, and publishing to Insight
• Fixed error when running Mosel models: "Couldn't reserve space for cygwin's heap"
• Fixed issue with copying a source file

• Fixed .Net issue with garbage collection of prob object while dependent objects are still in use
• Fixed .Net issue with garbage collector finalizing objects while native methods still need the resources they wrap

Fixed in Release 8.3.3

Xpress Mosel 4.6.1

• An issue has been fixed in the handling of extended types in packages

• Java: an issue has been fixed in the handling of connection string parameters

• objective modification now always reported to optimiser

• An issue has been fixed in the 'hex' driver when used for reading

• JSON reader: possible failure when loading a file containing only an empty string

• Using mmquad no longer prevents optimising linear objectives

• Fixed error when mmxnlp would crash if loaded without being licensed.

• Relocate IVE examples so they are visible to IVE's Complete Models wizard.

• Fixed long delays when saving files, running Mosel models, and publishing to Insight
• Fixed error when running Mosel models: "Couldn't reserve space for cygwin's heap"
• Copying a source file not working

Fixed in Release 8.3.2

• Fixed an issue in which XSLP_UNFINISHEDLIMIT was always used as value 1.

Fixed in Release 8.3.1

• Duplicate row eliminations should now be more robust against badly scaled rows.
• Fixed an issue with the loading of a basis file in compact format when super-basic variables are already present in the problem.
• Fixed a problem when turning crossover on for quadratic problems.
• Do not report the parallel crossover banner during MIP search.
• Fixed an uninitialized name bug in the Python interface.

• Fixed a potential problem with nonlinear eliminations in refinery type models.
• Improved numerical behavior for quadratic problems.

• Fixed a problem in which an infeasible converged solution was declared optimal.

• Fixed .Net issue with garbage collector finalizing objects while native methods still need the resources they wrap

• Remove some restrictions on number of active connections to license server.

New in Release 8.3

• The Xpress Python interface now allows for creating and solving nonlinear optimization problems. The modeling capabilities that were introduced with the first Python interface in Xpress 8.0 are extended to create constraints with expressions of arbitrary complexity using nonlinear operators. The interface also allows for accessing the full power of the Xpress-Nonlinear API for querying and modifying nonlinear problems.

• New function 'setname(mpvar,string)'
• New function 'newmuid' to generate a unique identifier
• New annotation property 'strict'
• New syntax for 'with' (equivalent to a single iteration 'forall' loop)

• Operator '@_' may receive now an entity with a reference count >1
• New function 'newmuid' to generate a unique identifier

• New compiler flag '-we' to handle warnings as errors

• New procedure 'getdsoprop' to retrieve module information
• Support for interval timers (routines settimer, gettimer and canceltimer)

• New functions 'fmin' and 'fmax'

• New function 'getact(robustctr)'

• xlsx/xls/csv: selecting an entire row for writing has the same effect as using the 'grow' option (same behavious as Excel)
• Excel driver: initialising a scalar with an empty cell now resets the variable (same behaviour as other drivers)

• New package for creating graphs in SVG format and displaying them in a web browser or within Workbench
• Graphing functionality includes circle, pie, polygon, polyline, xmltext objects and SVG styling for groups or individual objects
• This package supercedes the mmive module

• New procedure 'readsol' with an array to store the solution
• New function 'getname(nlctr)'

• Added support for server-side encryption using Amazon-managed encryption keys (SSE-S3)

• New module allowing Mosel to instantiate a Java virtual machine (Windows/Linux only)

• Xpress Workbench is the new integrated development environment for the Xpress Optimization Suite.
• Compile, run and debug Mosel models
• Deploy and remotely debug Xpress Insight apps.
• Syntax highlighting, autocompletion and code folding for Mosel and VDL source files and Xpress Insight companion files.
• Integrates with the Xpress tuner.
• Integrates with Tableau for FICO Server for Tableau workbook authoring

• Removed kalis parameters which are not prefixed with "kalis_" (they were already deprecated)
• Added a model "scenesalloc" in the examples to show the impact of symmetry breaking

Fixed in Release 8.3

• Non-empty annotation categories are now propagated in packages
• Annotation definitions are now always published properly in packages

• All drivers: an error is now reported when initialising a scalar with an unsufficient number of columns

• Includes all fixes applied since the initial Xpress 8.2 release with NonLinear 31.01.02 up to and including Xpress NonLinear 31.01.09. Please refer to the Xpress 8.2 patch notes for the specific fixes.

• Fixed a bug in the subroutine "disjunctive"
• Other minor bug fixes.

• Fixed Java issue when addMIPSol is called with a null name

New in Release 8.2

• It is now possible to tune a set of problem instances with the console Optimizer.
  • A set of problem instances can be specified by listing them in a text file and passing this file to the tune command as "tune probset <set-filename>".
  • Tuning target will be the aggregate across the set of instances.
• Setting TUNERTHREADS=-1 will now automatically use all available logical cores for tuning.
• The tuner output log now includes a summary of the importance of each individual control setting from the best control set.
• Crossover after a Barrier solve is now multi-threaded. A new control, CROSSOVERTHREADS, has been introduced to set the maximum number of threads to use during crossover.
• A new function, XPRSgetlastbarsol, has been introduced, which can be used to retrieve the solution from the last Barrier solve, assuming it was successful. This can e.g. be used for retrieving the solution from Barrier in case the crossover phase timed out.
• Efficiency changes to the multi-threaded MIP code:
  • Reduced the overhead for small, easy MIPs.
  • Reduced the memory usage for very large MIPs, especially those that are significantly reduced during the initial preprocessing.

• Option '-sdir' now accepts a list of directories and may be stated several times
• Debugger:
  • new 'breaksub' command
  • the model causing the interruption is now automatically selected
  • new command 'lsattr' to list types attributes
  • command 'print' now supports all type attributes (e.g. getsize(text))

• Loops (forall,repeat and while) can now be preceded by a label of the form 'string|identifier:'
• 'break' and 'next' accept now an identifier or string as parameter: it is interpreted as a label for the loop
• New constant INFINITY and functions 'isinf(real)', 'isfinite(real)'
• New function 'isdynamic'
• New option 'fctasproc': when it is used functions can be used as procedures (i.e. their return value is silently ignored)
• New procedure 'asproc' to call a function and ignore its return value
• User types (defining sets) can now be used to define array indices

• New option 'dbgbrksub'
• New command 'lsattr'

• Default compatility convention for modules and packages is changing:

        a1.b1.c1 is compatible with a2.b2.c2 if
        (a1=a2) and ((b1<b2) or ((b1=b2) and (c2>=c1)))

  Up to this release the rule was:

        (a1=a2) and (b1=b2) and (c2>=c1)

• New functions 'findtypeattr', 'getattr' and 'getnexttypeattr' for type attribute management

• New functions 'finddsofct','realtostr','hmfind','findtypeattr','getattr'
• New service 'SVC_REQTYPS' to list type requirements

• Support for HEAD requests (function 'httphead')
• New function 'httppending' to get a list of requests waiting for a reply

• New function 'waitexpired'
• 'tmp:' can now be used as the working directory of a remote model

• upgrade to SQLite 3.17.0

• xlsx: handling of ranges consisting in all columns or all rows now similar to Excel

• constructor 'text(x)' can now be used with structured entities

• New Mosel library for uploading/downloading objects from an Amazon S3 bucket

• Added support for UTF-8 strings

• Improved logic for finding initial points when none provided. New control XSLP_FINDIV indicating if initial values should automatically be determined
• Improved bound reduction in presolve through linear and quadratic coefficients. New control XSLP_LINQUADBR indicating if bound reduction should consider linear and quadratic coefficients as well as nonlinear expressions
• Support for variables with integer, semi-continuous or exploration type delta variables.
  • New delta types for nonlinear variables:
    • Integer delta: for variables that take values at integer multiples of a given step value
    • Semi-continuous delta: for variables that need to be changed at least by a given amount to have a measurable effect
    • Explore delta: for variables that are used in functions with zero partials at an unknown size
  • New API function XSLPchgdeltatype and Mosel function setdeltatype to mark variables as having a specific delta type
  • New control XSLP_GRIDHEURSELECT to indicate which heuristics to execute:
    • Enumeration: for small search spaces, evaluate all combinations
    • Cyclic: simple local search heuristics
    • Simulated annealing: simulated annealing enhanced local search heuristics
• Code generation tool for parallelizing user (black box) functions in Mosel. New Mosel function 'generateUFparallel' available in the mmxnlp package
• Extended capabilities for warm starting SLP solves using the LP barrier solver
  • Automatic fallback to the interior point solution found by barrier in case of crossover effort limits being hit
  • New control XSLP_BARSTARTOPS: allows fine tuning how barrier is applied in initial SLP iterations
  • New control XSLP_BARCROSSOVERSTART: specifies when to activate crossover; aimed at very large problem instances
  • New control XSLP_BARSTALLINGLIMIT: automatic numerical stalling fallback for barrier starts
  • New control XSLP_BARSTALLINGOBJLIMIT: automatic progress stalling fallback for barrier starts
  • New control XSLP_BARSTALLINGTOL: automatic progress stalling fallback tolerance for barrier starts

Fixed in Release 8.2

Xpress-Mosel 4.4.0

• An issue has been resolved in the handling of 'forall' loop constants
• Negative constants can now be used for indices in inline initialisation
• A constant set/list can now be initialised from a record field
• An issue has been resolved in the handling of '#line' markers in combination with empty statements in 'then' or 'else' constructs

• Windows: an event is now always available if 'wait(t)' returns before the time limit

• NULL data result now properly handled by 'SQLreadstring'
• An issue has been fixed in SQLexecute when it is used to export a set

• An issue has been fixed in the handling of comparisons of time/date/datetime when one of the operands is null
• 'findfiles' now ignores non-existent directories in filters
• 'findfiles' no longer includes a leading directory separator in file names of its result set
• 'findfiles' no longer fails to retrieve subdirectories on some file systems (Linux)

• Fixed non initialized error code when reading an empty set

• Fixed a possible memory leak when reaching memory exhaustion
• Improved tracking of allocated memory for .Net

New in Release 8.1

Xpress-Optimizer 30.01.04

• A new performance Tuner is available through the Optimizer console, the API and the Mosel modelling language. The Tuner automates the search for good control settings for solving a given problem, or class of problems, more efficiently.
  • Allows for tuning of all problem types supported by the Xpress Optimizer and Xpress Nonlinear.
  • Tuning can be performed against permuted versions of the original instance in order to ensure tuned results are more robust against small problem changes.
  • A new console command, TUNE, is available for starting a tuning session on a loaded problem.
  • A new API function, XPRStune, has been introduced for tuning a problem through the API.
  • New controls have been introduced for configuring the new Tuner:

    TUNERHISTORY     Whether to re-use previous results when restarting the Tuner.
    TUNERMAXTIME     Time limit for the entire tuning session.
    TUNERMETHOD      Selects which built-in method the Tuner should use when searching through possible control combinations.
    TUNERMODE        Can be used to specify that a Tuner session should be run before a problem is solved.
    TUNEROUTPUT      Controls the amount of logging output from the Tuner.
    TUNERPERMUTE     Number of permutations of the original problem to include in a Tuner session, in addition to the original problem.
    TUNERTARGET      The Tuner goal, such as improve solve time, best bound or best solution.
    TUNERTHREADS     Number of threads used by the Tuner
    TUNERMETHODFILE  The location of a file describing which control settings to tune against.
    TUNEROUTPUTPATH  A directory where Tuner results will be written to.
    TUNERSESSIONNAME String identifier for the Tuner session. Results will be recorded to a folder with this name in the TUNEROUTPUTPATH location.

• Dual Simplex now includes Sifting, which can speed up the solve of LPs with many more columns than rows. A new control, SIFTING, has been introduced to select when to run the Sifting algorithm.
• It is now possible to warm-start a Barrier solve by providing a primal solution, a dual solution or both. A starting solution can be loaded through the API with XPRSloadlpsol, or from a file using XPRSreadslxsol. Setting the control BARSTART=-1 will cause Barrier to use a pre-loaded solution for warm-starting, if one exists.
• A new local search heuristic is available for MIPs, based on interior point solutions. It is most useful for finding a first feasible solution. A new bit 6 has been introduced in HEURSEARCHROOTSELECT and HEURSEARCHTREESELECT to enable or disable this heuristic.
• Improved presolving for MIPs with a large number of precedence constraints. A new control, PREIMPLICATIONS, has been introduced to control this feature.
• A new control, MIPRAMPUP, has been introduced to control the ramp-up phase of parallel MIP.
• Aggregated Mixed Integer Rounding cuts have been improved for network-type problems.
• Flow path cuts are now disabled by default. To enable them, clear bit 11 of the CUTSELECT or the TREECUTSELECT controls.
• The Python API is now compatible with Python 2.7.
• The Python API now supports all Optimizer callback functions.
• The control CPUPLATFORM for selecting vectorization support now has two additional settings:

  -1: Automatically select the highest deterministic support [SSE2, AVX].
  -2: Automatically select the highest support [SSE2, AVX, AVX2].

  Selecting AVX2 can result in a different solve path than Generic, SSE2 or AVX.
• A new API function, XPRS_ge_setarchconsistency has been introduced. It can be used to ensure solver behavior that is consistent across systems with different levels of vectorization support [SSE2, AVX or AVX2].
• A new control, CROSSOVERITERLIMT, has been introduced to set the iteration limit for crossover.
• A new control, CROSSOVEROPS, has been introduced to allow for tuning of crossover.
• Added new control CROSSOVERITERLIMT to set iteration limit for crossover

• New option '-sdm sdm' to the the maximum depth of a stack dump. This can also be set with the environment variable MOSEL_SDMAX
• New execution option '-cov' and command 'cover[age]' to perform code coverage of models -
• New option '-sdir dir' to select the location of profiler/debugger source and report files
• Debugger: new command 'finish' (continue up to end of subroutine)

• New procedures 'fwrite[ln][_]' taking a stream number (integer) as the first argument. These procedures make it possible to write to an error stream
• Assertion error messages are now send to the model error stream
• New statement 'return' to quit the current subroutine
• New functions 'setmatherr', 'NAN' and 'isnan'
• New control parameter 'mathctrl' to disable failure on math error

• Improved efficiency of memory management
• Improved performance on writing text files in DOS format
• Operation R:=S with R being a range and S a set of integers containing one element is now supported
• Notation "\uxxxx" can now be used to represent Unicode characters in string constants
• After a runtime error Mosel now dumps its call stack. The maximum depth of this dump is defined using function 'XPRMsetsdmax' (default is 0)
• New functions 'XPRMsetsdmax' and 'XPRMgetsdmax'
• Function 'getmodprop(PROP_NAME)' now returns the initial model name. To get the internal unique name use property PROP_UNAME

• New service 'SVC_DSOSTRE' to implement a dso stream

• Verification of the syntax of annotations has been improved
• Procedure 'setparam' is now more restrictive regarding type conversion

• Support for documentation of annotations (doc.annot.*)

• New command 'covres' to retrieve code coverage information (model must have been run in profiling mode)
• New parameter 'sdmax' to set the maximum stack dump size
• Parameter 'dbgctrl': new value 'F' to continue execution up to the end of the current subroutine
• Command 'dbgflndx' now includes an additional array ('name') to report the function names. If used with no parameter it returns the line indices for all available routines. By default the result set is unsorted: option 'N' sorts it according to routine names and option 'L' uses the line indices as the sorting criteria
• Eval: new function 'getmodprop' to retrieve model information

• New option '-tc' to display the server configuration
• Configuration can now be updated while the server is running (Windows: with '-service reload' command; Unix: USR1 signal)
• New configuration variable 'CONFDIR': this is a directory path where are located additional configuration files (read after the main file)
• New context specific configuration variables RUN_BEFORE and RUN_AFTER
• New notation '[context]+': append definitions only if "context" exists
• New notation '[context]=': replace definitions "context"
• XPRMSRV_ACCESS: definition of a network from a host name (e.g. "myhost/24")
• XPRMSRV_ACCESS: new special identifier "SELF" (server hostname)

• New version of 'httpreply'/'httpreplycode' with an additional parameter to include supplementary headers to the response
• New procedure 'httpcancel' to cancel an asynchronous request
• New parameters 'http_maxcontime' and 'http_maxreqtime' to set maximum duration of requests
• Support for 'gzip' content encoding in results for httpget/put/post/del
• New function 'httpgetheader' to extract the header of a result file

• Propagate sdmax setting to remote instances
• New procedure 'getmodprop' to retrieve model information

• Support for control parameter 'mathctrl'

• Extented syntax now supports 3-digits parameter indices (e.g. '?101')
• Both extended and standard syntax can now be used in the same query
• The 'mmsqlite' driver now includes the extension functions from: http://www.sqlite.org/contrib/download/extension-functions.c

• LP format: append '/2' to quadratic obective
• Support for control parameter 'mathctrl'

• Upgrade to libxl 3.7.1

• Support for SSL operations with time limit (for https)

• New function 'formattext'

• New versions of 'getnode' & 'getnodes' accepting a 'text' parameter

• New optimisation option XPRS_TUNE to enable the tuner

• The automatic differentiation tool now supports parallel calculations. New control XSLP_CALCTHREADS can be used to manually set the number of threads used.
• The nonlinear eliminations in presolve have been made more efficient.
• The nonlinear presolver may now shrink the problem size. Use XSLP_PRESOLVESTATE to determine if the problem is in a presolved state. To prevent SLP from shrinking the problem size, set the XSLP_PRESOLVELEVEL control to smaller than full.
• XSLPgetslpsol always return the solution to the original problem. XSLPgetslpsol now only return values for columns and rows present in the original problem.
• New attributes XSLP_ORIGINALCOLS and XSLP_ORIGINALTOWS can be used to return the original number of model rows and columns in the problem.

• Add parameter KALIS_SCHEDULE_ENABLE_SHAVING to control whether or not activate shaving step when scheduling
• Fixed a bug in the tree search disjunction display in Xpress-IVE.
• Fixed a rare bug preventing Xpress-Kalis from finding the optimal solution to certain scheduling problems.

Fixed in Release 8.1

Xpress-Mosel 4.1

• An issue has been resolved in the processing of multi-lines annotations
• An issue has been resolved in the processing of annotations when option 'explterm' is being used
• An issue has been resolved in the handling of loop local variables of the form 'v=value' when 'value' is not an integer

• A potential memory corruption has been fixed in the 'findident' function when it is used by an NI routine during a debugging session
• The 'null:' IO driver no longer enforces binary mode (this was causing issues with XML output)
• Statement A+=B where both A and B are ranges and A is empty no longer results in a runtime error
• Option EP_STRIP no longer ignored (exportprob)

• 'urlencode' now handles UTF-8 characters properly

• An issue in the comparison to 'nullevent' has been fixed
• Event objects are now initialised with 'nullevent'

• A potential memory corruption has been fixed in the handling of empty results when reading sets or lists

• A potential 'bus error' has been fixed in 'getsize(text)' (Sparc only)

• JSON reader: notation '\uxxxx' can now process unicode values >255
• The default encoding of an XML document is now UTF-8

• A potential memory corruption has been fixed in the zip file driver

• Fixed the format of the eventual constant term in the objective when exporting a problem to LP file

New in Release 8.0.4

• The parallel MIP code has been completely rewritten to improve performance and scalability. The MIP API and solver interaction through callbacks remains mostly unaffected.
  • The worker tasks have been decoupled from the number of running threads. The solution path followed by the MIP solver now depends only on the worker task limit and not on the number of running solver threads.
  • A new control, MAXMIPTASKS, has been introduced to manually limit the number of MIP worker tasks. The number of running threads is still given by MIPTHREADS (which is derived from THREADS on automatic setting). By setting MAXMIPTASKS to a fixed number, the MIP solver will always return the same deterministic solution independent of the number of cores on the host machine, assuming that the LP threading controls have also been set to a fixed number. To guarantee a deterministic outcome for any number of cores, the following controls should be set: MAXMIPTASKS to a positive value. BARTHREADS to a positive value FORCEPARALLELDUAL=1 CONCURRENTTHREADS to a positive value, or do not run concurrent.
  • A new control MIPTERMINATIONMETHOD determines how to complete worker tasks in case of an early termination (time limit, node limit, etc).
  • Heuristics can now be run in parallel with cutting in deterministic mode. Previously, heuristics would only be run in parallel after cutting.
• It is now possible to create a basic solution from a not necessarily basic LP solution vector.
  • A new API function, XPRScrossoverlpsol, has been introduced to run the crossover algorithm on a user solution loaded through memory or from a file.
  • A new API function, XPRSloadlpsol, to load a solution vector from memory into the Optimizer. It will be set up as a solution without a basis, to the current LP problem in memory.
  • A new control, PREPROTECTDUAL, that can be used to specify whether the presolving of a problem with a loaded LP solution should avoid any reductions that could introduce dual infeasibilities.
• Introduced code support for AVX2 in the Barrier solver.
  • Set CPUPLATFORM=3 to enable AVX2 code support (when supported by the host CPU).
• Improved performance and numerical stability of crossover.
• A new control, BARORDERTHREADS, has been introduced to set the number of concurrent threads for the sparse matrix ordering in the barrier algorithm.
• New presolve reductions, strengthenings and reformulations for convex quadratic and conic problems.
  • A new control, PRECONEDECOMP, has been introduced to specify whether large cones should be decomposed when solving an MISOCP problem using an outer approximation method.
• Improved implementation of zero-half cutting for MIPs.
• The LP solution refiner can now run before postsolve to address any scaling infeasibilities. New bit option for XPRS_REFINEOPS has been added, and is enabled by default.
• The LP solution refiner now allows target values larger then the the feasibility and optimality tolerance, to be used on the post-solved problem.
• The effect of setting controls during the MIP search is now deterministic (if the MIP is solved in deterministic mode).
• A new control, HEURSEARCHROOTCUTFREQ, which can be used to specify how frequently root local search heuristics should be run between rounds of cuts.
• A new zero-objective local search heuristic to help find a first feasible solution for MIPs. Enabled by turning on bit 5 of the HEURSEARCHROOTSELECT and HEURSEARCHTREESELECT controls.
• A new control, ADDMIPSOLHEURISTIC, to control the local search heuristic that will be applied to infeasible or partial solutions loaded with XPRSaddmipsol.
• It is now possible to get rows and columns sorted as part of presolve. This could e.g. be used to ensure that a problem solves the same way regardless of the order in which variables and constraints have been declared.
  • A new control, PRESORT, has been introduced to specify whether to sort rows, columns or the global entity list.
  • New controls, PREPERMUTE and PREPERMUTESEED, have been introduce to allow rows, columns or the global entity list to be permuted randomly.
• The control MPSFORMAT now defaults to free format with a value of 1.
• The control CSTYLE no longer has any effect and will be removed from future releases.
• The control HEURSEARCHROOTSELECT new defaults to 53 which enables two additional local search heuristics during the root node processing.
• New Python interface for Xpress-Optimizer (requires Python 3.4 or newer and NumPy 1.1 or newer).
• The Java interface requires a minimum of Java 6 (internal version: 1.6.0)
• The Visual Basic interface is only supported via VB.NET

Mosel now represents all its internal strings in UTF-8 and supports character encoding conversions.

Remote Mosel instances can now be controled via remote queries in addition to usual XPRD/mmjobs queries. A query takes the form of a special file (to be open for reading) that is interpreted as a command: the data file returned by the instance is the result of the query. Supported commands:

  eval      evaluate an expression
  info      properties of Mosel, model/package/module
  lsloc     retrieve a list of local symbols
  lslib     list of available modules and packages
  profres   result of a profiling session
  dbglndx   mapping of line+file names in a model
  dbgbrkp   breakpoint management
  dbgstat   execution status
  dbgstlev  stack management
  dbgflndx  location of a subroutine

• New procedures 'publish' and 'unpublish' to control visibility of private and local symbols at runtime
• New functions _() and _c() for message translation
• New procedures write_ and writeln_ for message translation
• Annotations mc.msgdom and mc.msgid for message translation
• New procedure 'setioerr' to raise an IO error from the Mosel code

• New option '-l' to select the language
• Debugger: new command 'lslocal' to display a list of local symbols

• The standard protocol can now be disabled by setting the tcp port to -1 (in this case the server starts only if the ssh protocol is enabled)
• New server parameter SSH_CIPHERS (and command line option '-sc') to select which ciphers are accepted for SSH sessions

• The new 'xprnls' library implements various character encoding routines as well as message catalog management (for message translation)
• The 'xprnls' command tool is a character encoding converter (i.e. it makes it possible to convert text files from an encoding to another) and a message catalog processor

• Java: file manager routines now intercept all exceptions instead of only IOExceptions
• New function 'XPRDinstid' (C) and 'XPRDMosel.getId' (Java) to retrieve the ID of an instance (=node number)

• Java version: string handling functions accept now an additional parameter for selecting the character encoding (default: UTF-8)

• The Java interface requires a minimum of Java 6 (internal version: 1.6.0)
• New method to retrieve local identifiers (XPRMModel.localIdentifiers)
• When a PrintStream is used as a text output stream via the 'mmjava:' driver and no encoding is specified, data is sent as text strings to the 'print' method (this enables Java encoding conversion).

• The Visual Basic interface is no longer supported by VB6, only via VBA and VB.NET
• VBA: some functions that take or return pointers have been changed to use Variant rather than Long (in 32-bit) or LongPtr (in 64-bit)
• VBA: The XPRM_IO_CB function can now be used to create callback-based I/O handlers for any text output, not just as an argument to XPRMsetdefstream

• All routines working with strings now expect UTF-8 strings
• Source file encoding may be specified using the '!@encoding' annotation to be put at the beginning of the file. For instance: !@encoding:ISO-8859-15
• New compiler option '-x' to generate a POT file from a model (message translation)
• Verification of the syntax of annotations has been improved

• All routines working with strings now expect UTF-8 strings
• New 'enc:' filename prefix to select encoding. By default text files are expected to be in UTF-8 and detect UTF-16 and UTF-32.
• 'initialisations' data files can specify an encoding using the special comment '!@encoding' to be put at the beginning of the file. For instance: !@encoding:CP1252
• New function 'XPRMdbg_getnextlocal' to enumerate local symbols during a debugging session
• New function 'XPRMdbg_getlndx' to retrieve all lines indices at once
• If restriction 'NoExec' is active, modules that are not located in read-only directories are now ignored
• Mosel can now be installed under bin32/lib32/dso32 (32bit build) and bin64/lib64/dso64 (64bit build). This makes it possible to have both 32bit and 64bit distributions under the same directory tree
• Operation R:=S with R being a range and S a set of integers containing one element is now supported

• Deprecated IO routine 'IOCTRL_ERROR' no longer supported
• IO driver routine 'IOCTRL_OPEN' now receives an additional parameter (encoding of the stream)

• Switch to xprnls for message translation

• Consecutive requests to a given server are now processed using a single connection
• New control parameters 'http_expire' and 'http_maxconn' to manage the pool of connections
• Switch to UTF-8 encoding

• Deleting the file "shmem:*" has the effect of deleting all memory files handled by 'shmem'
• 'rmt:' driver: r/w error messages from the remote host are now displayed
• New parameter 'defaultnode': this is the node number used by 'rmt:' when no node reference is given - switch to UTF-8 encoding

• Switch to xprnls for message translation

• Communication with the database in UTF-8
• Option 'debug' now disabled if model is not compiled in debug mode

• Support for ANSI (default on Posix) and UNICODE (default for Windows) ODBC interfaces
• Interface can be selected with the 'enc:' filename prefix
• Option 'debug' now disabled if model is not compiled in debug mode
• New procedure SQLindices to retrieve the indices of a table
• SQLite driver: support for ':memory:' and empty file name databases

• Switch to xprnls for message translation

• Switch to xprnls for message translation

• Switch to UTF-8 encoding
• Driver 'csv': strings starting with '0' or spaces are no longer implicitly interpreted as numbers
• Empty records are now skipped when reading a list of records

• Switch to UTF-8 encoding

• All string operations now performed in UTF-8
• New value -1 for 'qtype' (quoting mode) disables quoting
• 'zip' and 'tar' routines can select the encoding used for file names (in the archive) using the 'enc:' prefix on the archive name
• IMCI: new functions 'gettime'/'settime', 'getdate'/'setdate', 'getdatetime'/'setdatetime'

• XML documents are now encoded/decoded according to the specified encoding
• JSON documents are now properly handled as UTF-8
• Procedure 'getnodes' and 'getnode' no longer fail when applied to node 0 on an empty document
• New function 'getsize(xmldoc)' to get the number of nodes of a document
• Control parameter 'xml_keeputf' has been removed (no longer necessary)

• Switch to UTF-8 encoding

• Support for UTF-8 in file names

• Switch to UTF-8 encoding
• Support for program parameters containing non-ascii symbols

• Added support for restricted R on FAC

• New default convergence mechanism for improved solution quality
  • new control XSLP_VALIDATIONTARGET_K to specify target first order optimality accuracy.
  • new control XSLP_VALIDATIONTARGET_R to specify target relative validation accuracy.
  • new values for XSLP_CONVERGENCEOPS to allow disabling / enabling the target convergence options.
  • new default value for XSLP_CONVERGENCEOPS to reflect the use of the target convergence checks.
  • the traditional SLP convergence controls now default to -1, their values are dynamically adjusted during the run. U ser set values are respected as normal and are not modified. The controls that now default to -1 are XSLP_CTOL, XSLP_ATOL_A, XSLP_ATOL_R, XSLP_MTOL_A, XSLP_MTOL_R, XSLP_ITOL_A, XSLP_ITOL_R, XSLP_STOL_A, XSLP_STOL_R, XSLP_MVTOL, XSLP_XTOL_A, XSLP_XTOL_R, XSLP_OTOL_A, XSLP_OTOL_R, XSLP_VTOL_A, XSLP_VTOL_R, XSLP_EVTOL_A, XSLP_EVTOL_R, XSLP_ECFTOL_A, XSLP_ECFTOL_R, XSLP_WTOL_A and XSLP_WTOL_R.
• The mixed integer nonlinear solver MISLP has been reworked
  • much improved general performance.
  • the default MINLP solver now runs in parallel mode for the tree search.
  • greatly increased heuristics capability for the MINLP solver.
• Probing algorithm based on the bound reduction algorithm to further tighten variable bounds; it works on several levels (which can be set automatically or by the user) for applying probing to different classes of variables.
• The VB interface is no longer supported, except in VB.NET. User functions written in VB6 or VBA are still supported, provided that they do not need to call the XSLP API.
• Upgraded Knitro to version 10.0.

• Added C/C++/Java/.NET functions to iterate over constraints and constraint terms
• Added/extended XPRBterm and XPRMqterm classes to Java and .Net to allow inspection of iterated constraint terms
• Changed getIIS Java/.Net method to retrieve the first IIS by default instead of the IIS approximation
• Added error code return to C++ functions that can fail on out of memory
• The VB interface is no longer supported, except in VB.NET
• The interface to Java requires a minimum of Java 6 (internal version: 1.6.0)

• Supports Mosel UTF-8 encoding

• Supports Mosel UTF-8 encoding

Fixed in Release 8.0.4

• Debugger: 'getsize' now reports the expected value when applied to strings produced from external types

• Redeclaration of a constant scalar no longer leads to a corrupted bim file
• An issue has been resolved in the processing of multi-lines annotations

• 64bit platforms: a potential memory corruption has been fixed

• Excel driver: a possible crash on Windows 8.1 64Bit and newer has been fixed
• xls/xlsx/csv drivers: conversion from boolean to string now valid

• Parsetext now updates 'endparse' when used with a parse context
• mmsystem.text driver now fails when trying to write to an unitialised object

• Optimiser status 'MIP_LP_NOT_OPTIMAL' now interpreted as 'unfinished'
• 'defsecurevecs' no longer ignores set of rows if a set of variables is used
• A potential memory corruption has been fixed in getloaded{mpvars|linctrs}
• 'savemipsol' no longer reports an error when used on a presolved problem

• Fixed issue with 'getIIS' when used to access the IIS approximation
• Fixed 'printcut' to print using problem pointer
• Fixed error message in 'printctr'/'printidxset'
• Fixed extra cut printing in 'xbgetcuts'

• Fixed Excel issue, where model deployed to VBA would occasionally have incorrect working directory
• Fixed issue where file might fail to laod if start and end aligned with memory boundary

Known issues in Release 8.0