feat: wlroots 0.20.2, and the two packages the pango line was waiting on - #318
Merged
Conversation
THE COMPOSITOR GATE IS OPEN. Everything wlroots needs was already here — wayland-server/client/protocols/scanner/egl, libdrm, pixman, xkbcommon, libinput, libudev, libseat, egl, glesv2, gbm, libdisplay-info — and this is the package that spends them. wlroots.wlroots 0.20.2 fork, six generators in build.mcpp, no sh, no python compat.pcre2 10.44 glib's GRegex is written against it compat.fribidi 1.0.16 the bidi algorithm pango needs for RTL text WHY TWO OF THE THREE ARE DESCRIPTORS AND ONE IS A FORK The criterion is generators, not line count. pcre2 is 133k lines and needs no fork: the release tarball ships pcre2.h.generic, config.h.generic and pcre2_chartables.c.dist precisely so a build without autotools is possible. fribidi generates seven UCD tables with eight C programs — and ships the output, verified one for one against the seven `#include "*.tab.i"` in its sources. wlroots has six generators and ships none of their output.⚠️ import wlroots; IS NOT A CONVENIENCE wlroots' 121 public headers contain no `extern "C"`, and two are not valid C++ at all: five declarations use C99 `const float color[static 4]`, and three struct members are named `namespace`, `delete` and `class`. A C++ consumer cannot include them under any arrangement of `extern "C"`. The fork writes C++-safe copies — `[static N]` reduced to `[N]`, keyword members given an `#ifdef __cplusplus` spelling at the same offset with the `#else` arm upstream verbatim — and the module wraps the lot. WHAT THE TESTS ASSERT, AND WHY NOT "IT LINKED" Each check names a specific decision, so a failure says which one broke: wlroots WLR_HAS_* is exactly the default set; wlr_box_intersection computes 60,20 40x30; wlr_scene_rect_set_color is callable (the rewritten [static 4]); namespace_ is reachable from C++; the PINNED pnp.ids resolves ACR to Acer and DEL to Dell — and "JJJ", whose 15-bit key no entry occupies, returns NULL. Not an arbitrary byte triple: PNP_ID keeps the low five bits of each character, so "\x01\x02\x03" packs to the same key as "ABC" and returns AboCom System Inc. pcre2 \p{Han} and \X prove SUPPORT_UNICODE; `.` matching one code point with PCRE2_UTF and one byte without proves UTF-8 mode; and byte 0xE9 NOT being alphabetic proves the C-locale character tables, i.e. that the build did not run pcre2_dftables against a machine locale. fribidi three different bidi types from one table; a Hebrew run placed first and reversed in an RTL paragraph; "(" mirroring to ")" while "a" mirrors to nothing; ALEF right-joining and BEH dual-joining. Also: the wayland-protocols example moves to 1.49.1 and now static_asserts on the enum headers that release added. Verified on both toolchains from clean trees. wlroots' own CI is green on both legs; its fork release and all three CN mirrors are published.
CI on this PR failed with wlr_layer_shell_v1.h:73: use of enum 'zwlr_layer_surface_v1_keyboard_interactivity' without previous declaration from a build that passed locally. The shim that renames the `namespace` argument — layer-shell is the only one of 133 generated protocol headers that is not valid C++ — reached the scanner's real output with `#include_next`, and that is an include which can SUCCEED and deliver the wrong file. A missing file would have said so. The fork now names the file by absolute path (mcpplibs/wlroots, CI green on both toolchains) and 0.20.2 is re-cut. Safe here and only here: the descriptor had never been published, so nothing in the world had extracted it. CN URL moves to the `0.20.2-1` container: gitcode refuses to replace an asset of the same name in an existing release, so the corrected bytes needed a new one. Verified — its sha256 equals the GLOBAL tarball's.
The previous tarball still put wayland-scanner's headers in the build program's out dir, which is appended AFTER every dependency's include directory. In CI — never locally, across four differently-cleaned trees, both mcpp releases and both standalone and workspace invocations — that produced wlr_output_power_management_v1.h:39: use of enum 'zwlr_output_power_v1_mode' without previous declaration from a header that includes its protocol header on line 5. An include that SUCCEEDED and delivered the wrong file: the failure a search path can have. Every generated header now goes to the directory the manifest names FIRST, which is where config.h already was and for the same reason. Fork CI green on both toolchains. CN container moves to 0.20.2-2: gitcode will not replace an asset of the same name, so each corrected tarball needs a new container. Only the last is referenced.
Four CI rounds, one cause, never reproducible locally: wlr_layer_shell_v1.h:73: use of enum 'zwlr_layer_surface_v1_keyboard_interactivity' without previous declaration `mcpp::action` schedules a generated .c before the compile that needs it, because mcpp fixes the source set during prepare. It cannot do the same for a HEADER: nothing declares a dependency on one, and the MODULE INTERFACE UNIT is compiled first since every other TU needs its BMI. So the module was compiled against files the action had created and not yet filled — an include that SUCCEEDED and delivered nothing. A missing file says so; this never did, which is finally what identified it. wayland-scanner now runs inside build.mcpp via posix_spawn (no shell), so every generated header exists before any compiler starts, and each invocation is checked for spawn, exit status AND a non-empty output. Fork CI green on both toolchains; verified here against the published tarball with every store cleared. Also adds §19 to the design doc: the compositor gate, the GObject stack, the four 'include succeeded and gave the wrong file' failures, and the measured boundary of what is left (gdbus-codegen, 8351 lines, blocking gio and therefore pango).
They cannot be verified in THIS PR, and the reason is structural rather than a missing step: gnome.glib depends on compat.pcre2, which this PR is what publishes. An example redirects ONE namespace at a package's index — mcpp registers at most one project index repo — so `gnome` points at this checkout and `compat.pcre2` is looked up in the published index, where it does not exist yet. wlroots now passes here (all twelve assertions, including the pinned pnp.ids resolving ACR to Acer), so this PR is complete on its own. glib, gobject and gmodule 2.82.5 are built, tested on both toolchains, released and mirrored; their descriptors and examples land as soon as compat.pcre2 is on the index.
macOS and Windows failed while both Linux legs passed, and the cause was in this PR rather than in any package: the 1.49.1 enum-header includes were appended at the END of the file, past the `#endif` that closes `#ifdef __linux__`. So they compiled on platforms where the dependency is never fetched — the example gates it with `cfg(linux)` — and the scan failed before anything was built. Appending to a file is not the same as adding to it. Moved above the `#else`, where the rest of the test lives; verified it still passes on Linux.
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
1 participant
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.
The compositor gate is open. Everything wlroots needs was already in this index — wayland-server/client/protocols/scanner/egl, libdrm, pixman, xkbcommon, libinput, libudev, libseat, egl, glesv2, gbm, libdisplay-info. This is the package that spends them.
wlroots.wlrootsbuild.mcpp, nosh, nopythoncompat.pcre2GRegexis written against itcompat.fribidiWhy two are descriptors and one is a fork
The criterion is generators, not line count. pcre2 is 133k lines and needs no fork: the release tarball ships
pcre2.h.generic,config.h.genericandpcre2_chartables.c.distprecisely so a build without autotools is possible. fribidi generates seven UCD tables with eight C programs — and ships the output, verified one for one against the seven#include "*.tab.i"in its sources.wlroots has six generators and ships none of their output.
import wlroots;is not a convenience — it is the only way inwlroots' 121 public headers contain no
extern "C", and two of them are not valid C++ at all:g++ says
expected primary-expression before 'static', and then — the parse never recovering — reports every later declaration as "has not been declared", which sends you hunting for missing feature guards. Three struct members are also namednamespace,deleteandclass.The fork writes C++-safe copies:
[static N]→[N], keyword members given an#ifdef __cplusplusspelling at the same offset with the#elsearm upstream verbatim. Same layout, same ABI, one spelling per language.What the tests assert
Each check names a specific decision, so a failure says which one broke rather than "the package is broken":
WLR_HAS_*is exactly the default set;wlr_box_intersectioncomputes60,20 40x30;wlr_scene_rect_set_coloris callable (the rewritten[static 4]);namespace_is reachable from C++; the pinnedpnp.idsresolves ACR→Acer and DEL→Dell, and"JJJ"returns NULL. Not an arbitrary byte triple —PNP_IDkeeps the low five bits of each character, so"\x01\x02\x03"packs to the same key as"ABC"and returns AboCom System Inc.\p{Han}and\XproveSUPPORT_UNICODE;.matching one code point withPCRE2_UTFand one byte without proves UTF-8 mode; and byte 0xE9 not being alphabetic proves the C-locale character tables, i.e. that the build did not runpcre2_dftablesagainst a machine locale.(mirroring to)whileamirrors to nothing; ALEF right-joining and BEH dual-joining.The
wayland-protocolsexample moves to 1.49.1 andstatic_asserts on the enum headers that release added.Verification
Both toolchains, clean trees. wlroots' own CI is green on both legs (run); the fork release and all three CN mirrors are published.