dev-build/gn
GN is a meta-build system that generates build files for Ninja
Screenshots
View
Download
BrowseChangeLog
commit d7d76a6adf12c315c4743550c44c96d307936613
Author: lucascouts <lucascs@protonmail.com>
Date: Mon Sep 7 16:34:38 2026 -0300
clean(dev-build/gn): drop gn-gen r5 and r7, two revisions behind the r8 in use
The ebuild applies gn-gen-r8.patch. Revision-suffixed patches accumulate one per rebase and nobody deletes the predecessor. Found by scripts/gentoo-parity.sh.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
commit eada4f2dfda6dffdefb6af85ab85ade35c1704d2
Author: lucascouts <lucascs@protonmail.com>
Date: Mon Sep 7 11:12:25 2026 -0300
fix(dev-util/vulkan-tools, sys-apps/flatpak): two build/test breakages ::gentoo already patches, del(4 orphan patches), mod(8 ebuilds): tag the last PATCHES rows - ALIGN reaches zero
The last thirteen open findings: the eleven remaining PATCHES rows and the two
UNDOCUMENTED ones. With these the sweep reports 222 rows, ALL JUSTIFIED, and
exits 0 for the first time.
TWO REAL BREAKAGES.
dev-util/vulkan-tools is missing #include <ctime> in vulkaninfo.cpp and fails
to build against libc++ 23. The timing is the whole story and was measured
against the GitHub API, not guessed: EGIT_COMMIT here is 532e2c0e from
2026-09-04T02:24Z, and upstream's fix 462d9819 landed at 18:35Z the SAME DAY -
sixteen hours later. The snapshot missed it by one working day. The patch
carries a delete-me note naming the commit that obsoletes it, so eapply will
fail loudly at the next bump rather than rot silently.
sys-apps/flatpak, both ebuilds: tests/make-test-runtime.sh copies
/usr/lib/locale/C.* with no guard, so building the test runtime dies where that
glob matches nothing. ::gentoo adds "|| true"; checked both 1.18.2 and 1.19.0
tarballs and neither has it.
Both patches were renamed version-agnostic on the way in, which promptly created
three fresh UNDOCUMENTED rows - the comparator sees a different filename on each
side and cannot know the content matches. Tagged as such. Worth remembering:
copying a patch from ::gentoo does not close a PATCHES row, it changes what the
row says.
FOUR ORPHAN PATCHES DELETED, which is what the two UNDOCUMENTED rows were.
tdesktop-6.9.3-{no-dispatch,qt6-no-wayland,use-pkgconfig-for-cmark-gfm}.patch
and nodejs-24.11.1-gcc15-ppc.patch: no ebuild references any of them, and the
versions they name are long gone (telegram-desktop is at 7.2.5, nodejs at 24.20
and 26.8). They are recoverable from git if that judgement was wrong.
EIGHT TAGS, each measured by reverse-applying ::gentoo's patch against the
tarball this overlay actually ships:
liborcus, gn, networkmanager-openvpn reverse-apply cleanly - upstream took
the change, ::gentoo's is a backport
bleachbit upstream went FURTHER: 6.0.3's locale
regex also accepts three-digit region
codes the Gentoo patch does not
imath the patch edits src/python/config/
ModuleDefine.cmake and Imath 3.2 has no
src/ at all
vim-core ::gentoo applies six Gentoo distro
patches pinned at VIM_PATCHES_VERSION
9.1.1432, four hundred patchlevels
behind the 9.2.1037 shipped here
TWO OF THE EIGHT LEAVE SOMETHING OPEN, and say so in the tag rather than
implying it is settled:
sentry-native ::gentoo's no-fuzz-test patch does NOT apply - hunk 2 fails
against 0.16.5. So sentry_fuzz_json is still built and
registered under USE=test here, and their reason for disabling
it stands. Rebasing the patch is the fix.
xdg-desktop-portal their sandbox-disable-failing-tests patch does not apply
either, both hunks failing against 1.22.1. test_dynamiclauncher
.py and test_location.py are still listed and still want
pipewire, network and /dev/fuse. Their optional-gstreamer patch
is a different matter and must stay absent - it invents a USE
flag upstream does not have.
Verified. pkgcheck against a worktree at 6abe6fb56 is identical on all twelve
packages touched. Tarballs came from DISTDIR where cached and from upstream
otherwise - gitlab.com, github.com, download.gnome.org, distfiles.obentoo.org.
--self-test 24/24, shellcheck -S warning clean.
A20 re-measured 224 -> 222 with a note on what changes now: until today a rising
ALIGN count was ordinary, and from here any ALIGN at all is a NEW divergence
that arrived after 2026-09-07. The sweep is finally usable as a gate.
Four md5-cache orphans went with the egencache run: vim-core-9.2.1036,
bleachbit-6.0.2 and flatpak-1.18.1, none with an ebuild since earlier bumps.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
commit e54848aada11d20d41adb6617d628784d88fa0ce
Author: lucascouts <lucascs@protonmail.com>
Date: Mon Aug 3 00:39:36 2026 -0300
mod(dev-build/gn-0.2501): pina o commit que gerou o snapshot self-hosted
O overlay gera e hospeda o proprio tarball deste pacote, entao ele deixa de
ser consumidor de um artefato e passa a ser o publicador dele. A receita
documentada no ebuild dizia `git archive ... HEAD`, e HEAD anda: regenerar o
arquivo num outro dia publicava BYTES DIFERENTES SOB O MESMO NOME. O commit
produtor sobrevivia so na mensagem de cda533be e no pax header do tarball --
nenhum dos dois alcancavel a partir do ebuild que o usuario compila.
Declara GN_COMMIT=3fd3b0624d8cba16927853600130b2c33d4e7928 e reescreve a
receita em funcao dele. A variavel e' load-bearing, nao decorativa: o
src_configure sintetiza last_commit_position.h (o tarball nao carrega .git,
por isso o --no-last-commit-position), e agora emite a mesma string que o
build/gen.py:266-275 do upstream produz -- "<num> (<sha12>)" -- no lugar do
PV cru. Um pin errado aparece em `gn --version`; nao passa em silencio.
antes: #define LAST_COMMIT_POSITION "0.2501"
depois: #define LAST_COMMIT_POSITION "2501 (3fd3b0624d8c)"
Evidencia executada, nao afirmada:
- receita rodada verbatim a partir do commit pinado, num clone limpo de
gn.googlesource.com, e `cmp` contra o arquivo que distfiles.obentoo.org
realmente serve: exit 0, byte a byte. SHA512 bate com o Manifest.
- contraprova do defeito: a MESMA receita com HEAD gera 887292 bytes contra
os 887192 publicados. O upstream ja andou para 17b0057970fa (2502) desde
o dia do bump, entao a divergencia e' de hoje, nao hipotetica.
- pkgcheck scan --cache=-git --exit error dev-build/gn -> EXIT=0
(unico achado e' PythonCompatUpdate para python3_15, nivel warning)
- emerge -pv dev-build/gn -> resolve, EXIT=0
- md5-cache regenerado (egencache --repositories-configuration): _md5_
0048321107d6f865a131f28573326ff9 -> beb63090e669a9777357ebcdc3c62230
O nome do arquivo ainda NAO codifica o commit. A alternativa considerada era
gn-$-$.tar.xz, que tornaria a colisao impossivel, mas
exige re-upload manual para distfiles.obentoo.org -- e commitar o rename
antes do upload deixaria o SRC_URI em 404, exatamente o que o hold = true do
record existe para evitar. Decisao do mantenedor: adotar no proximo bump,
que sobe um tarball novo de qualquer forma.
Ref: story 006, R7.1/R7.2/R7.4, tarefa 7.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
commit cda533becd05de1b1c63ef3eff4f711f6b4d80de
Author: lucascouts <lucascs@protonmail.com>
Date: Sun Aug 2 17:37:02 2026 -0300
add(metadata/md5-cache/dev-build/gn-0.2501), mod(.autoupdate/packages.toml), up(dev-build/gn-0.2374-r1 -> 0.2501)
O bentoo passa a gerar e hospedar o próprio snapshot do gn.
Upstream publica commits, nunca releases, então alguém precisa empacotar.
Quem fazia isso era o ::gentoo, em deps.gentoo.zip — e aquele espelho só
carrega a versão em que o ::gentoo está, o que prendia este overlay ao
0.2374 (o gn-0.2501.tar.xz de lá dá 404 até hoje). O tarball agora é gerado
aqui e publicado em distfiles.obentoo.org.
A receita foi PROVADA byte a byte antes de ser adotada: reproduzi o
gn-0.2374.tar.xz que o deps.gentoo.zip publica, a partir do clone oficial,
com "git archive --format=tar --prefix=gn-0.2374/ <commit> | xz -9e", e o
diff -rq contra o publicado deu idêntico. O arquivo agora servido foi
rebaixado e conferido: sha512 igual ao gerado, content-type
application/x-xz, 1175 entradas.
PV = 0.<count> de "git describe HEAD --abbrev=12 --match initial-commit",
ou seja 0.2501 em 3fd3b0624d8c. São 127 commits desde o 0.2374.
files/gn-0.2374-test-iwyu.patch REMOVIDO: o upstream o incorporou em
07ea23e5, o primeiro commit após o 0.2374 — o dry-run acusa "previously
applied", isto é, mantê-lo quebraria o src_prepare. O gn-gen-r8.patch segue
aplicando (só offsets, sem fuzz).
Validado com build real, não apenas com emerge -pv: ebuild
unpack/prepare/configure/compile sob FEATURES=network-sandbox, gcc 16.1.1.
O upstream passou a exigir C++23 em 88604adb e o log confirma -std=c++23
sem um único erro de compilação; o binário resultante roda e reporta 0.2501
(a síntese do last_commit_position.h está correta). A fase install falha no
dobin por chown sem root — limitação do host, não do ebuild; pkg_setup e
merge real ficam por verificar.
Record com hold = true: um bump aqui nunca é troca de PV, o tarball tem que
ser gerado e publicado antes. A probe apontando para o deps.gentoo.zip foi
mantida com o sentido invertido — quando ela reportar versão >= a nossa, o
::gentoo alcançou e a auto-hospedagem pode ser aposentada.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
commit 6b1397bc6ea90b3f21bdf81b6d01f4b7ba18ba0f
Author: lucascouts <lucascs@protonmail.com>
Date: Thu May 7 19:09:16 2026 -0300
up(dev-build/gn-0.2374 -> 0.2374-r1)
commit 87f4e676aff60f7cf21fd1c959d94be0d8b85e8a
Author: lucascouts <lucascs@protonmail.com>
Date: Thu May 7 19:07:11 2026 -0300
add(dev-build/gn-0.2374)
Author: lucascouts <lucascs@protonmail.com>
Date: Mon Sep 7 16:34:38 2026 -0300
clean(dev-build/gn): drop gn-gen r5 and r7, two revisions behind the r8 in use
The ebuild applies gn-gen-r8.patch. Revision-suffixed patches accumulate one per rebase and nobody deletes the predecessor. Found by scripts/gentoo-parity.sh.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
commit eada4f2dfda6dffdefb6af85ab85ade35c1704d2
Author: lucascouts <lucascs@protonmail.com>
Date: Mon Sep 7 11:12:25 2026 -0300
fix(dev-util/vulkan-tools, sys-apps/flatpak): two build/test breakages ::gentoo already patches, del(4 orphan patches), mod(8 ebuilds): tag the last PATCHES rows - ALIGN reaches zero
The last thirteen open findings: the eleven remaining PATCHES rows and the two
UNDOCUMENTED ones. With these the sweep reports 222 rows, ALL JUSTIFIED, and
exits 0 for the first time.
TWO REAL BREAKAGES.
dev-util/vulkan-tools is missing #include <ctime> in vulkaninfo.cpp and fails
to build against libc++ 23. The timing is the whole story and was measured
against the GitHub API, not guessed: EGIT_COMMIT here is 532e2c0e from
2026-09-04T02:24Z, and upstream's fix 462d9819 landed at 18:35Z the SAME DAY -
sixteen hours later. The snapshot missed it by one working day. The patch
carries a delete-me note naming the commit that obsoletes it, so eapply will
fail loudly at the next bump rather than rot silently.
sys-apps/flatpak, both ebuilds: tests/make-test-runtime.sh copies
/usr/lib/locale/C.* with no guard, so building the test runtime dies where that
glob matches nothing. ::gentoo adds "|| true"; checked both 1.18.2 and 1.19.0
tarballs and neither has it.
Both patches were renamed version-agnostic on the way in, which promptly created
three fresh UNDOCUMENTED rows - the comparator sees a different filename on each
side and cannot know the content matches. Tagged as such. Worth remembering:
copying a patch from ::gentoo does not close a PATCHES row, it changes what the
row says.
FOUR ORPHAN PATCHES DELETED, which is what the two UNDOCUMENTED rows were.
tdesktop-6.9.3-{no-dispatch,qt6-no-wayland,use-pkgconfig-for-cmark-gfm}.patch
and nodejs-24.11.1-gcc15-ppc.patch: no ebuild references any of them, and the
versions they name are long gone (telegram-desktop is at 7.2.5, nodejs at 24.20
and 26.8). They are recoverable from git if that judgement was wrong.
EIGHT TAGS, each measured by reverse-applying ::gentoo's patch against the
tarball this overlay actually ships:
liborcus, gn, networkmanager-openvpn reverse-apply cleanly - upstream took
the change, ::gentoo's is a backport
bleachbit upstream went FURTHER: 6.0.3's locale
regex also accepts three-digit region
codes the Gentoo patch does not
imath the patch edits src/python/config/
ModuleDefine.cmake and Imath 3.2 has no
src/ at all
vim-core ::gentoo applies six Gentoo distro
patches pinned at VIM_PATCHES_VERSION
9.1.1432, four hundred patchlevels
behind the 9.2.1037 shipped here
TWO OF THE EIGHT LEAVE SOMETHING OPEN, and say so in the tag rather than
implying it is settled:
sentry-native ::gentoo's no-fuzz-test patch does NOT apply - hunk 2 fails
against 0.16.5. So sentry_fuzz_json is still built and
registered under USE=test here, and their reason for disabling
it stands. Rebasing the patch is the fix.
xdg-desktop-portal their sandbox-disable-failing-tests patch does not apply
either, both hunks failing against 1.22.1. test_dynamiclauncher
.py and test_location.py are still listed and still want
pipewire, network and /dev/fuse. Their optional-gstreamer patch
is a different matter and must stay absent - it invents a USE
flag upstream does not have.
Verified. pkgcheck against a worktree at 6abe6fb56 is identical on all twelve
packages touched. Tarballs came from DISTDIR where cached and from upstream
otherwise - gitlab.com, github.com, download.gnome.org, distfiles.obentoo.org.
--self-test 24/24, shellcheck -S warning clean.
A20 re-measured 224 -> 222 with a note on what changes now: until today a rising
ALIGN count was ordinary, and from here any ALIGN at all is a NEW divergence
that arrived after 2026-09-07. The sweep is finally usable as a gate.
Four md5-cache orphans went with the egencache run: vim-core-9.2.1036,
bleachbit-6.0.2 and flatpak-1.18.1, none with an ebuild since earlier bumps.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
commit e54848aada11d20d41adb6617d628784d88fa0ce
Author: lucascouts <lucascs@protonmail.com>
Date: Mon Aug 3 00:39:36 2026 -0300
mod(dev-build/gn-0.2501): pina o commit que gerou o snapshot self-hosted
O overlay gera e hospeda o proprio tarball deste pacote, entao ele deixa de
ser consumidor de um artefato e passa a ser o publicador dele. A receita
documentada no ebuild dizia `git archive ... HEAD`, e HEAD anda: regenerar o
arquivo num outro dia publicava BYTES DIFERENTES SOB O MESMO NOME. O commit
produtor sobrevivia so na mensagem de cda533be e no pax header do tarball --
nenhum dos dois alcancavel a partir do ebuild que o usuario compila.
Declara GN_COMMIT=3fd3b0624d8cba16927853600130b2c33d4e7928 e reescreve a
receita em funcao dele. A variavel e' load-bearing, nao decorativa: o
src_configure sintetiza last_commit_position.h (o tarball nao carrega .git,
por isso o --no-last-commit-position), e agora emite a mesma string que o
build/gen.py:266-275 do upstream produz -- "<num> (<sha12>)" -- no lugar do
PV cru. Um pin errado aparece em `gn --version`; nao passa em silencio.
antes: #define LAST_COMMIT_POSITION "0.2501"
depois: #define LAST_COMMIT_POSITION "2501 (3fd3b0624d8c)"
Evidencia executada, nao afirmada:
- receita rodada verbatim a partir do commit pinado, num clone limpo de
gn.googlesource.com, e `cmp` contra o arquivo que distfiles.obentoo.org
realmente serve: exit 0, byte a byte. SHA512 bate com o Manifest.
- contraprova do defeito: a MESMA receita com HEAD gera 887292 bytes contra
os 887192 publicados. O upstream ja andou para 17b0057970fa (2502) desde
o dia do bump, entao a divergencia e' de hoje, nao hipotetica.
- pkgcheck scan --cache=-git --exit error dev-build/gn -> EXIT=0
(unico achado e' PythonCompatUpdate para python3_15, nivel warning)
- emerge -pv dev-build/gn -> resolve, EXIT=0
- md5-cache regenerado (egencache --repositories-configuration): _md5_
0048321107d6f865a131f28573326ff9 -> beb63090e669a9777357ebcdc3c62230
O nome do arquivo ainda NAO codifica o commit. A alternativa considerada era
gn-$-$.tar.xz, que tornaria a colisao impossivel, mas
exige re-upload manual para distfiles.obentoo.org -- e commitar o rename
antes do upload deixaria o SRC_URI em 404, exatamente o que o hold = true do
record existe para evitar. Decisao do mantenedor: adotar no proximo bump,
que sobe um tarball novo de qualquer forma.
Ref: story 006, R7.1/R7.2/R7.4, tarefa 7.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
commit cda533becd05de1b1c63ef3eff4f711f6b4d80de
Author: lucascouts <lucascs@protonmail.com>
Date: Sun Aug 2 17:37:02 2026 -0300
add(metadata/md5-cache/dev-build/gn-0.2501), mod(.autoupdate/packages.toml), up(dev-build/gn-0.2374-r1 -> 0.2501)
O bentoo passa a gerar e hospedar o próprio snapshot do gn.
Upstream publica commits, nunca releases, então alguém precisa empacotar.
Quem fazia isso era o ::gentoo, em deps.gentoo.zip — e aquele espelho só
carrega a versão em que o ::gentoo está, o que prendia este overlay ao
0.2374 (o gn-0.2501.tar.xz de lá dá 404 até hoje). O tarball agora é gerado
aqui e publicado em distfiles.obentoo.org.
A receita foi PROVADA byte a byte antes de ser adotada: reproduzi o
gn-0.2374.tar.xz que o deps.gentoo.zip publica, a partir do clone oficial,
com "git archive --format=tar --prefix=gn-0.2374/ <commit> | xz -9e", e o
diff -rq contra o publicado deu idêntico. O arquivo agora servido foi
rebaixado e conferido: sha512 igual ao gerado, content-type
application/x-xz, 1175 entradas.
PV = 0.<count> de "git describe HEAD --abbrev=12 --match initial-commit",
ou seja 0.2501 em 3fd3b0624d8c. São 127 commits desde o 0.2374.
files/gn-0.2374-test-iwyu.patch REMOVIDO: o upstream o incorporou em
07ea23e5, o primeiro commit após o 0.2374 — o dry-run acusa "previously
applied", isto é, mantê-lo quebraria o src_prepare. O gn-gen-r8.patch segue
aplicando (só offsets, sem fuzz).
Validado com build real, não apenas com emerge -pv: ebuild
unpack/prepare/configure/compile sob FEATURES=network-sandbox, gcc 16.1.1.
O upstream passou a exigir C++23 em 88604adb e o log confirma -std=c++23
sem um único erro de compilação; o binário resultante roda e reporta 0.2501
(a síntese do last_commit_position.h está correta). A fase install falha no
dobin por chown sem root — limitação do host, não do ebuild; pkg_setup e
merge real ficam por verificar.
Record com hold = true: um bump aqui nunca é troca de PV, o tarball tem que
ser gerado e publicado antes. A probe apontando para o deps.gentoo.zip foi
mantida com o sentido invertido — quando ela reportar versão >= a nossa, o
::gentoo alcançou e a auto-hospedagem pode ser aposentada.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
commit 6b1397bc6ea90b3f21bdf81b6d01f4b7ba18ba0f
Author: lucascouts <lucascs@protonmail.com>
Date: Thu May 7 19:09:16 2026 -0300
up(dev-build/gn-0.2374 -> 0.2374-r1)
commit 87f4e676aff60f7cf21fd1c959d94be0d8b85e8a
Author: lucascouts <lucascs@protonmail.com>
Date: Thu May 7 19:07:11 2026 -0300
add(dev-build/gn-0.2374)