Repository navigation
win: Fix standalone installer MSVC runtime and build the installer in CI - #664
Open
wenzeslaus wants to merge 28 commits into
Open
wenzeslaus wants to merge 28 commits into
wenzeslaus wants to merge 28 commits into
Conversation
…stall The msvcrt2019 archive published on download.osgeo.org changed to msvcrt2019-14.2-2.tar.bz2; update the archive name and the download and unpacked sizes reported by the installer to match it. Run g.mkfontcap in a hidden section placed after the Microsoft runtime section so that the runtime DLLs it needs are already installed when it runs (OSGeo#5663).
Build the NSIS-based standalone installer and upload it as an artifact so that changes to the installer can be tested without a local MSYS2/OSGeo4W setup. The job mirrors the flow used by wingrass-maintenance-scripts: build with the OSGeo4W scripts, make install into the OSGeo4W tree with the packaging fixups from mswindows/osgeo4w/package.sh, run GRASS-Packager.bat, and compile GRASS-Installer.nsi with the AccessControl and untgz NSIS plugins. The workflow runs only for changes under mswindows or to the workflow itself, and on manual dispatch. Written with the assistance of Claude Code.
The OSGeo4W CI build script removes etc/fontcap from the dist tree so that the test tree regenerates it on first run, but make install requires the file and has no rule to rebuild it from the top-level Makefile. Rebuild it with the g.mkfontcap Makefile before make install. Also set PYTHONHOME like the build script does so the OSGeo4W Python used by make install finds its standard library.
Owner
|
The install went ok. I have some screenshots, but some are missing because I mistakenly made them inside the trimmed down Windows Sandbox, which didn't save them. Screenshots: 
|
Owner
|
I checked the box to open GRASS after the installation is done on the installer. It didn't open grass. I double-clicked on the shortcut, and it didn't open grass either. I called the same thing that the shortcut does inside a terminal, and I saw the following: C:\Users\WDAGUtilityAccount>"C:\Program Files\GRASS 8.6\grass86.bat" --gui
Traceback (most recent call last):
File "C:\Program Files\GRASS 8.6\etc\grass86.py", line 2537, in <module>
main()
File "C:\Program Files\GRASS 8.6\etc\grass86.py", line 2178, in main
find_grass_python_package()
File "C:\Program Files\GRASS 8.6\etc\grass86.py", line 2127, in find_grass_python_package
raise RuntimeError(msg)
RuntimeError: \c\OSGeo4W\apps\grass\grass86\etc\python with the grass Python package does not exist. Is the installation of GRASS complete?We don't have a "C:\OSGeo4W" folder either, the standalone instal installs in "C:\Program Files\GRASS 8.6". The install seems fine, and contain the grass package: |
|
Modified and now Grass correctly starts. |
…er, making grassXY.py able to start Co-authored-by: Antonio Galea <antonio.galea@gmail.com>
echoix
reviewed
Jul 14, 2026
Migrates the r.null testsuite from gunittest to pytest. Everything goes through grass.tools (r.describe results read with .json, no gs.parse_command). The three checks from the old testsuite are preserved: setnull on an integer map with -i, setnull on a floating point map with -f, and filling NULL cells with the null parameter.
CodeQL Code Quality had some AI findings on a recently changed file, a test file for g.md2man. The tests suggested seemed legitimate, and they are passing. I am thus submitting them here, after testing them in my fork. Original text explaining the findings: ------- The test asserts that there are exactly 3 bullet list markers for the input `"- first\n- second\n - nested\n"`, but this input only has 3 items (first, second, nested). The assertion `== 3` may be correct, but it should be verified that the nested item also produces the same `.IP \(bu 4n` marker, since nested items could produce a different marker format. If the nested list intentionally uses the same marker, this is fine, but the test name `test_unordered_list_nested` suggests the nesting behavior is what's being tested, and there is no assertion verifying that the nested item is actually rendered as nested (e.g., inside `.RS`/`.RE` blocks relative to the parent items). Consider adding an assertion that checks the order of `.RS`, `.IP`, and `.RE` to confirm correct nesting structure. ---------- The `test_comment_parsing_unit` test covers multi-line and inline comments, but there is no unit test for `strip_html_comments` behavior inside a fenced code block (i.e., that a comment inside a fence is preserved). The integration test `test_comment_kept_in_fenced_code` covers this via subprocess, but a direct unit test of `strip_html_comments` with a fenced block containing a comment would provide more precise coverage of that branch in `gmd.py`. * Apply suggested fix to utils/g.md2man/tests/g_md2man_test.py from Copilot Autofix Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com> * Apply suggested fix to utils/g.md2man/tests/g_md2man_test.py from Copilot Autofix Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com> * Apply ruff format --------- Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com>
Fix the today config type, skip PLY lexer docstrings in autodoc, remove the unsupported logo theme option, correct the toctree entries, and make the sample data links work again. The initial changes were AI-assisted (OpenCode with DeepSeek), the review and follow-up fixes used Claude Code (with Fable). --------- Co-authored-by: Edouard Choinière <27212526+echoix@users.noreply.github.com> Co-authored-by: Vaclav Petras <wenzeslaus@gmail.com>
…OSGeo#7672) * set sqlite as default backend on init * read tgis_db_version from globals if no DB is initialized * add tests for STDS construction
…OSGeo#7337) * move to Tools API * add upper/lower filter range options) (backwards compatible) for both vector and raster based filtering * propagate options of underlying tools (r.clump) * add tests Co-authored-by: Markus Neteler <markus@neteler.org> * Edit html * Unchange header line * typos --------- Co-authored-by: Edouard Choinière <27212526+echoix@users.noreply.github.com> Co-authored-by: Markus Neteler <markus@neteler.org>
…e function" (OSGeo#7673) Follow-up from OSGeo#7660 This PR applies the same logic as the other PR, where how sscanf can also return EOF "if input failure occurs before the first receiving argument was assigned". (See https://en.cppreference.com/c/io/fscanf). Here, the little touch that is different for `raster/r.out.vtk/writeascii.c` and `raster3d/r3.out.vtk/writeVTKData.c`, is that I assume that if there's the `EOF` case, or any invalid case for the `null_value`, we would still want the null value to be 0. I think that's the design intention behind the code. This was assisted by AI (Copilot-generated fix suggestions for code scanning alerts), but we can reason through it and I applied some myself in order to batch them in one single PR. This should fix: - https://github.com/OSGeo/grass/security/code-scanning/525 - https://github.com/OSGeo/grass/security/code-scanning/526 - https://github.com/OSGeo/grass/security/code-scanning/527 - https://github.com/OSGeo/grass/security/code-scanning/528 - https://github.com/OSGeo/grass/security/code-scanning/531 - https://github.com/OSGeo/grass/security/code-scanning/532 - https://github.com/OSGeo/grass/security/code-scanning/533 * Potential fix for code scanning alert no. 525: Incorrect return-value check for a 'scanf'-like function Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com> * Potential fix for code scanning alert no. 528: Incorrect return-value check for a 'scanf'-like function Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com> * Update writeascii.c * r3.out.vtk: Fix "Incorrect return-value check for a 'scanf'-like function" for null value * r.watershed: Fix "Incorrect return-value check for a 'scanf'-like function" for disturbed_land default value Adresses https://github.com/OSGeo/grass/security/code-scanning/532 --------- Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
We observed some random, transient timeouts on different tests. For one of them, t.list, there was no reason to fail on macOS, where there are 3 cores instead of 2 in GitHub Actions. There's a possibility that there's a problem with the temporal library locking up somewhere, but there’s also a small chance that the r.mapcalc calls are now using multiple processors. It "costs" not much to add this env var so that the number of threads is limited, even if it doesn't fix all problems; it might just fix some potential ones, since we are already running in parallel in that section.
AI findings on a recently modified file found a typo in a translated string. This adds "be" in "Values should be in between"... Apply suggested fix to raster3d/r3.out.vtk/writeVTKData.c from Copilot Autofix Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com>
The ctypes interface assigned the number of columns to the rows key and the number of rows to the cols key, for raster as well as 3D raster maps. The swapped values are stored in the temporal database when a map is registered and are reported by t.info, which therefore disagreed with r.info and r3.info. The number of cells was not affected, because it is the product of both. Maps registered before this fix keep the swapped values in the temporal database until they are registered again. Update the doctests which asserted the swapped values and add a test comparing the metadata read through the interface with the output of r.info and r3.info. The test was written with the help of Claude Code.
Python 3.10 receives security fixes only and reaches end of life in October 2026, before the next release. Per RFC 8, raise the minimum supported Python version from 3.10 to 3.11. Update the version in REQUIREMENTS.md, pyproject.toml (requires-python, Black, Pylint, Pyright, and isort) and CI (pytest matrix and MIN_PYTHON_VERSION). Update places in code which add code for the older version: - utils/mkdocs.py: datetime.fromisoformat parses the Z suffix used by recent Git natively since 3.11, so the fallback is no longer needed. - Use the datetime.UTC alias added in 3.11. Ruff enables this check (UP017) once it infers the higher target version from requires-python. - Drop the toml extra for bandit. It provided a TOML parser for reading the bandit configuration from pyproject.toml, which the standard library covers since 3.11 with tomllib. Here the deciding version is the one the workflow runs, which is well past 3.11 already.
* CMake: Build man pages from markdown Generate the markdown documentation (--md-description + mkmarkdown.py) in all three doc paths (modules, GUI, scripts) and convert man pages from it with g.md2man instead of from HTML with g.html2man, mirroring the Autotools build. Install g.md2man.py and gmd.py, and install the generated docs/mkdocs tree, which was previously not installed at all. Add man/build_manpages.py, a sweep run after the index scripts that converts any markdown page without an up-to-date man page. It ports man/Makefile's MANPAGES wildcard to CMake, which has no build-time glob, and produces the index and dynamically named topic_* man pages that the CMake build previously did not ship. Run the doc generators in the tool source directory like Autotools does. This fixes the SOURCE CODE footer in all CMake-built pages, which previously pointed to a nonexistent build-directory path in the repository and fell back to an Accessed date instead of the last commit, and makes the copy-source-into-build-dir steps unnecessary. Also fix build_class.py receiving the documentation directory as its year argument. Verified against an Autotools build: g.md2man output for the same generated markdown is byte-identical. Generated with AI assistance (Claude Fable), including the build verification against the Autotools doc tree. * address review
The table connection raised a new ImportError with the hint as its message, so the message of the original error, which says which module is missing, was replaced by the hint. Add the hint as a note to the original error instead. From the Python documentation of BaseException.add_note(): Add the string note to the exception's notes which appear in the standard traceback after the exception string. The traceback then shows both the missing module and the hint. Exception notes need Python 3.11, which is now the minimum version.
The test loader raised a new ImportError to say which module could not be imported. The message used e.message, which does not exist in Python 3, so the handler raised AttributeError instead of the intended error. Add the module name as a note to the original error instead. From the Python documentation of BaseException.add_note(): Add the string note to the exception's notes which appear in the standard traceback after the exception string. The traceback then shows both the import error and the module name. Exception notes need Python 3.11, which is now the minimum version.
) An empty WHERE result left an empty data array, which the classification algorithms then indexed at data[0] and data[count - 1]. The values printed as class breaks were whatever the heap happened to hold, which is zeroed on Linux but not on Windows, making the test flaky there. Warn the user and keep the zero-initialized breaks when nothing is selected, and reject an empty array in AS_class_apply_algorithm.
…tor geometry computation (OSGeo#7329)
The Scatterplot, Timeline, and Temporal Plot tools raised a new ImportError with the hint as its message, so the message of the original error, which says which module is actually missing, was replaced by the hint. Add the hint as a note to the original error instead. From the Python documentation of BaseException.add_note(): Add the string note to the exception's notes which appear in the standard traceback after the exception string. The traceback then shows both the missing module and the hint. The consumer code (various places) shows always its own message, so the not, not the original message is not used. These changes do not change that. Exception notes need Python 3.11, which is now the minimum version.
…7697) locale.getencoding() was added in Python 3.11, so the fallbacks to the deprecated locale.getdefaultlocale() are never used now that 3.11 is the minimal supported version. Call locale.getencoding() directly. The two places calling locale.getlocale() get the same treatment. That function is available in all supported versions, so those fallbacks could not be reached even before.
echoix
force-pushed
the
fix-msvc-runtime-dll-sizes
branch
from
July 14, 2026 17:26
5771ab1 to
ea22bb2
Compare
…aking grassXY.py able to start Co-authored-by: Antonio Galea <antonio.galea@gmail.com>
… installer, making grassXY.py able to start" This reverts commit 0b77028.
|
Just tested latest installer - confirmed working. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This fixes the standalone Windows installer and adds a CI workflow that builds it, so the fix (and future installer changes) can be tested from CI artifacts.
Installer fix (mswindows/GRASS-Installer.nsi.tmpl)
msvcrt2019-14.2-2.tar.bz2as published on download.osgeo.org, and correct the download size (920 KB, verified against the server) and unpacked size shown by the installer.g.mkfontcapin a new hidden-RunMkFontCapsection placed after the Microsoft runtime section, so the runtime DLLs it needs are already installed when it runs. Previously it ran at the end of the GRASS section, before the runtime download, which is part of [Bug] install of WinGRASS-8.5dev-a277d8547c-18-Setup VCRUNTIME140.dll not found error OSGeo/grass#5663.CI workflow (.github/workflows/windows-installer.yml)
New workflow that builds the NSIS standalone installer and uploads
WinGRASS-*-Setup.exeas an artifact (7-day retention). It mirrors the production flow of wingrass-maintenance-scripts:mswindows/osgeo4w/build_osgeo4w.shdirectly to skip theUNITTESTadditions that would otherwise end up in the package.make installintoC:\OSGeo4W\apps\grass\grassXYplus the packaging fixups frommswindows/osgeo4w/package.sh(config.h switch,etc/grass.py->etc/grassXY.pyrename that the generated launcher depends on, MinGW runtime DLLs copied into the packagebin). The MSVC.libgeneration is skipped as it is only needed for the OSGeo4W package.GRASS-Packager.bat, followed by a content sanity check (xcopy failures do not stop the batch file).makensiswith the AccessControl and untgz plugins (not bundled with NSIS; downloaded with pinned SHA256 checksums, unicode DLLs installed intoPlugins\x86-unicodeas described in the Trac wiki).The workflow triggers on changes under
mswindows/**, changes to the workflow itself, and manual dispatch. Note that themakensisstep uses the solid LZMA compression from the installer script and may take a while on a hosted runner.The CI workflow was written with the assistance of Claude Code.
🤖 Generated with Claude Code