Pango 1.56.1 for mcpp — CN mirror

文件	最后提交记录	最后更新时间
.github	Pango 1.56.1 for mcpp — three members, each exposing a C++23 module The package the index said was blocked. gnome.gio landed, so the gate is open: PangoFontMap declares G_IMPLEMENT_INTERFACE(G_TYPE_LIST_MODEL, …), and the type cannot even register without it. The test asserts that at RUN TIME rather than taking the build's word for it. gnome.pango itemisation, bidi, line breaking, markup 851 exports gnome.pangoft2 fontconfig picks the file, FreeType draws 88 gnome.pangocairo the cairo backend 284 WHY A FORK: generators, not line count. Pango has three and this adds a fourth (the module wrapper), so the tree carries no sh and no python. ⚠️ config.h is the odd one: upstream writes configure_file(output: 'config.h') with NO input template, so meson emits a #define per key and there is nothing to substitute into. The file is WRITTEN, and every value in it is a decision. Two are arithmetic — PANGO_BINARY_AGE = minor*100+micro, PANGO_INTERFACE_AGE = minor odd ? 0 : micro — and pango_version_check reads the first, so a wrong value makes a correct comparison answer wrongly at run time. ⭐ ONE TEST PRODUCES PIXELS. pangocairo renders "Hello 世界" to an ARGB32 surface and counts non-transparent pixels: 1,492 here. Reaching that costs seven packages, and a blank image means one of them is not doing its job. It degrades HONESTLY when the runner has no fonts — that case is reported, not passed over, because "0 families so the check did not run" and "the text rendered" must not look alike. ⚠️ A SILENT DEGRADATION THIS FORK WALKED INTO. HAVE_CAIRO_FREETYPE was dropped from config.h by an editing slip. Nothing failed to build. pangocairo simply registered NO backends and died at run time: Pango-CRITICAL: Unknown $PANGOCAIRO_BACKEND value. Available backends are: <- an empty list then segfaulted. The test now reads the font map's FONT TYPE and requires CAIRO_FONT_TYPE_FT, which is the assertion that catches it immediately. ⚠️ freedesktop.cairo's module is NOT sufficient on its own — measured on 1.18.2: 470 names, ZERO enumerators (no CAIRO_FORMAT_ARGB32, no CAIRO_FONT_TYPE_FT) and no cairo_t. The index's own cairo example does not notice because it writes BOTH `#include <cairo.h>` and `import freedesktop.cairo;`. pangocairo cannot mix the routes — pangocairo.h reaches glib and therefore <stdio.h>, so `struct _IO_FILE` becomes two entities — so gnome.pangocairo scans cairo.h itself. That line can go when cairo is fixed. Six tests (three members × two routes), green on gcc 16.1 and llvm 22.1.	1 个月前
mcpp	Pango 1.56.1 for mcpp — three members, each exposing a C++23 module The package the index said was blocked. gnome.gio landed, so the gate is open: PangoFontMap declares G_IMPLEMENT_INTERFACE(G_TYPE_LIST_MODEL, …), and the type cannot even register without it. The test asserts that at RUN TIME rather than taking the build's word for it. gnome.pango itemisation, bidi, line breaking, markup 851 exports gnome.pangoft2 fontconfig picks the file, FreeType draws 88 gnome.pangocairo the cairo backend 284 WHY A FORK: generators, not line count. Pango has three and this adds a fourth (the module wrapper), so the tree carries no sh and no python. ⚠️ config.h is the odd one: upstream writes configure_file(output: 'config.h') with NO input template, so meson emits a #define per key and there is nothing to substitute into. The file is WRITTEN, and every value in it is a decision. Two are arithmetic — PANGO_BINARY_AGE = minor*100+micro, PANGO_INTERFACE_AGE = minor odd ? 0 : micro — and pango_version_check reads the first, so a wrong value makes a correct comparison answer wrongly at run time. ⭐ ONE TEST PRODUCES PIXELS. pangocairo renders "Hello 世界" to an ARGB32 surface and counts non-transparent pixels: 1,492 here. Reaching that costs seven packages, and a blank image means one of them is not doing its job. It degrades HONESTLY when the runner has no fonts — that case is reported, not passed over, because "0 families so the check did not run" and "the text rendered" must not look alike. ⚠️ A SILENT DEGRADATION THIS FORK WALKED INTO. HAVE_CAIRO_FREETYPE was dropped from config.h by an editing slip. Nothing failed to build. pangocairo simply registered NO backends and died at run time: Pango-CRITICAL: Unknown $PANGOCAIRO_BACKEND value. Available backends are: <- an empty list then segfaulted. The test now reads the font map's FONT TYPE and requires CAIRO_FONT_TYPE_FT, which is the assertion that catches it immediately. ⚠️ freedesktop.cairo's module is NOT sufficient on its own — measured on 1.18.2: 470 names, ZERO enumerators (no CAIRO_FORMAT_ARGB32, no CAIRO_FONT_TYPE_FT) and no cairo_t. The index's own cairo example does not notice because it writes BOTH `#include <cairo.h>` and `import freedesktop.cairo;`. pangocairo cannot mix the routes — pangocairo.h reaches glib and therefore <stdio.h>, so `struct _IO_FILE` becomes two entities — so gnome.pangocairo scans cairo.h itself. That line can go when cairo is fixed. Six tests (three members × two routes), green on gcc 16.1 and llvm 22.1.	1 个月前
upstream	Pango 1.56.1 for mcpp — three members, each exposing a C++23 module The package the index said was blocked. gnome.gio landed, so the gate is open: PangoFontMap declares G_IMPLEMENT_INTERFACE(G_TYPE_LIST_MODEL, …), and the type cannot even register without it. The test asserts that at RUN TIME rather than taking the build's word for it. gnome.pango itemisation, bidi, line breaking, markup 851 exports gnome.pangoft2 fontconfig picks the file, FreeType draws 88 gnome.pangocairo the cairo backend 284 WHY A FORK: generators, not line count. Pango has three and this adds a fourth (the module wrapper), so the tree carries no sh and no python. ⚠️ config.h is the odd one: upstream writes configure_file(output: 'config.h') with NO input template, so meson emits a #define per key and there is nothing to substitute into. The file is WRITTEN, and every value in it is a decision. Two are arithmetic — PANGO_BINARY_AGE = minor*100+micro, PANGO_INTERFACE_AGE = minor odd ? 0 : micro — and pango_version_check reads the first, so a wrong value makes a correct comparison answer wrongly at run time. ⭐ ONE TEST PRODUCES PIXELS. pangocairo renders "Hello 世界" to an ARGB32 surface and counts non-transparent pixels: 1,492 here. Reaching that costs seven packages, and a blank image means one of them is not doing its job. It degrades HONESTLY when the runner has no fonts — that case is reported, not passed over, because "0 families so the check did not run" and "the text rendered" must not look alike. ⚠️ A SILENT DEGRADATION THIS FORK WALKED INTO. HAVE_CAIRO_FREETYPE was dropped from config.h by an editing slip. Nothing failed to build. pangocairo simply registered NO backends and died at run time: Pango-CRITICAL: Unknown $PANGOCAIRO_BACKEND value. Available backends are: <- an empty list then segfaulted. The test now reads the font map's FONT TYPE and requires CAIRO_FONT_TYPE_FT, which is the assertion that catches it immediately. ⚠️ freedesktop.cairo's module is NOT sufficient on its own — measured on 1.18.2: 470 names, ZERO enumerators (no CAIRO_FORMAT_ARGB32, no CAIRO_FONT_TYPE_FT) and no cairo_t. The index's own cairo example does not notice because it writes BOTH `#include <cairo.h>` and `import freedesktop.cairo;`. pangocairo cannot mix the routes — pangocairo.h reaches glib and therefore <stdio.h>, so `struct _IO_FILE` becomes two entities — so gnome.pangocairo scans cairo.h itself. That line can go when cairo is fixed. Six tests (three members × two routes), green on gcc 16.1 and llvm 22.1.	1 个月前
.gitignore	Pango 1.56.1 for mcpp — three members, each exposing a C++23 module The package the index said was blocked. gnome.gio landed, so the gate is open: PangoFontMap declares G_IMPLEMENT_INTERFACE(G_TYPE_LIST_MODEL, …), and the type cannot even register without it. The test asserts that at RUN TIME rather than taking the build's word for it. gnome.pango itemisation, bidi, line breaking, markup 851 exports gnome.pangoft2 fontconfig picks the file, FreeType draws 88 gnome.pangocairo the cairo backend 284 WHY A FORK: generators, not line count. Pango has three and this adds a fourth (the module wrapper), so the tree carries no sh and no python. ⚠️ config.h is the odd one: upstream writes configure_file(output: 'config.h') with NO input template, so meson emits a #define per key and there is nothing to substitute into. The file is WRITTEN, and every value in it is a decision. Two are arithmetic — PANGO_BINARY_AGE = minor*100+micro, PANGO_INTERFACE_AGE = minor odd ? 0 : micro — and pango_version_check reads the first, so a wrong value makes a correct comparison answer wrongly at run time. ⭐ ONE TEST PRODUCES PIXELS. pangocairo renders "Hello 世界" to an ARGB32 surface and counts non-transparent pixels: 1,492 here. Reaching that costs seven packages, and a blank image means one of them is not doing its job. It degrades HONESTLY when the runner has no fonts — that case is reported, not passed over, because "0 families so the check did not run" and "the text rendered" must not look alike. ⚠️ A SILENT DEGRADATION THIS FORK WALKED INTO. HAVE_CAIRO_FREETYPE was dropped from config.h by an editing slip. Nothing failed to build. pangocairo simply registered NO backends and died at run time: Pango-CRITICAL: Unknown $PANGOCAIRO_BACKEND value. Available backends are: <- an empty list then segfaulted. The test now reads the font map's FONT TYPE and requires CAIRO_FONT_TYPE_FT, which is the assertion that catches it immediately. ⚠️ freedesktop.cairo's module is NOT sufficient on its own — measured on 1.18.2: 470 names, ZERO enumerators (no CAIRO_FORMAT_ARGB32, no CAIRO_FONT_TYPE_FT) and no cairo_t. The index's own cairo example does not notice because it writes BOTH `#include <cairo.h>` and `import freedesktop.cairo;`. pangocairo cannot mix the routes — pangocairo.h reaches glib and therefore <stdio.h>, so `struct _IO_FILE` becomes two entities — so gnome.pangocairo scans cairo.h itself. That line can go when cairo is fixed. Six tests (three members × two routes), green on gcc 16.1 and llvm 22.1.	1 个月前
README.mcpp.md	Pango 1.56.1 for mcpp — three members, each exposing a C++23 module The package the index said was blocked. gnome.gio landed, so the gate is open: PangoFontMap declares G_IMPLEMENT_INTERFACE(G_TYPE_LIST_MODEL, …), and the type cannot even register without it. The test asserts that at RUN TIME rather than taking the build's word for it. gnome.pango itemisation, bidi, line breaking, markup 851 exports gnome.pangoft2 fontconfig picks the file, FreeType draws 88 gnome.pangocairo the cairo backend 284 WHY A FORK: generators, not line count. Pango has three and this adds a fourth (the module wrapper), so the tree carries no sh and no python. ⚠️ config.h is the odd one: upstream writes configure_file(output: 'config.h') with NO input template, so meson emits a #define per key and there is nothing to substitute into. The file is WRITTEN, and every value in it is a decision. Two are arithmetic — PANGO_BINARY_AGE = minor*100+micro, PANGO_INTERFACE_AGE = minor odd ? 0 : micro — and pango_version_check reads the first, so a wrong value makes a correct comparison answer wrongly at run time. ⭐ ONE TEST PRODUCES PIXELS. pangocairo renders "Hello 世界" to an ARGB32 surface and counts non-transparent pixels: 1,492 here. Reaching that costs seven packages, and a blank image means one of them is not doing its job. It degrades HONESTLY when the runner has no fonts — that case is reported, not passed over, because "0 families so the check did not run" and "the text rendered" must not look alike. ⚠️ A SILENT DEGRADATION THIS FORK WALKED INTO. HAVE_CAIRO_FREETYPE was dropped from config.h by an editing slip. Nothing failed to build. pangocairo simply registered NO backends and died at run time: Pango-CRITICAL: Unknown $PANGOCAIRO_BACKEND value. Available backends are: <- an empty list then segfaulted. The test now reads the font map's FONT TYPE and requires CAIRO_FONT_TYPE_FT, which is the assertion that catches it immediately. ⚠️ freedesktop.cairo's module is NOT sufficient on its own — measured on 1.18.2: 470 names, ZERO enumerators (no CAIRO_FORMAT_ARGB32, no CAIRO_FONT_TYPE_FT) and no cairo_t. The index's own cairo example does not notice because it writes BOTH `#include <cairo.h>` and `import freedesktop.cairo;`. pangocairo cannot mix the routes — pangocairo.h reaches glib and therefore <stdio.h>, so `struct _IO_FILE` becomes two entities — so gnome.pangocairo scans cairo.h itself. That line can go when cairo is fixed. Six tests (three members × two routes), green on gcc 16.1 and llvm 22.1.	1 个月前
mcpp.toml	Pango 1.56.1 for mcpp — three members, each exposing a C++23 module The package the index said was blocked. gnome.gio landed, so the gate is open: PangoFontMap declares G_IMPLEMENT_INTERFACE(G_TYPE_LIST_MODEL, …), and the type cannot even register without it. The test asserts that at RUN TIME rather than taking the build's word for it. gnome.pango itemisation, bidi, line breaking, markup 851 exports gnome.pangoft2 fontconfig picks the file, FreeType draws 88 gnome.pangocairo the cairo backend 284 WHY A FORK: generators, not line count. Pango has three and this adds a fourth (the module wrapper), so the tree carries no sh and no python. ⚠️ config.h is the odd one: upstream writes configure_file(output: 'config.h') with NO input template, so meson emits a #define per key and there is nothing to substitute into. The file is WRITTEN, and every value in it is a decision. Two are arithmetic — PANGO_BINARY_AGE = minor*100+micro, PANGO_INTERFACE_AGE = minor odd ? 0 : micro — and pango_version_check reads the first, so a wrong value makes a correct comparison answer wrongly at run time. ⭐ ONE TEST PRODUCES PIXELS. pangocairo renders "Hello 世界" to an ARGB32 surface and counts non-transparent pixels: 1,492 here. Reaching that costs seven packages, and a blank image means one of them is not doing its job. It degrades HONESTLY when the runner has no fonts — that case is reported, not passed over, because "0 families so the check did not run" and "the text rendered" must not look alike. ⚠️ A SILENT DEGRADATION THIS FORK WALKED INTO. HAVE_CAIRO_FREETYPE was dropped from config.h by an editing slip. Nothing failed to build. pangocairo simply registered NO backends and died at run time: Pango-CRITICAL: Unknown $PANGOCAIRO_BACKEND value. Available backends are: <- an empty list then segfaulted. The test now reads the font map's FONT TYPE and requires CAIRO_FONT_TYPE_FT, which is the assertion that catches it immediately. ⚠️ freedesktop.cairo's module is NOT sufficient on its own — measured on 1.18.2: 470 names, ZERO enumerators (no CAIRO_FORMAT_ARGB32, no CAIRO_FONT_TYPE_FT) and no cairo_t. The index's own cairo example does not notice because it writes BOTH `#include <cairo.h>` and `import freedesktop.cairo;`. pangocairo cannot mix the routes — pangocairo.h reaches glib and therefore <stdio.h>, so `struct _IO_FILE` becomes two entities — so gnome.pangocairo scans cairo.h itself. That line can go when cairo is fixed. Six tests (three members × two routes), green on gcc 16.1 and llvm 22.1.	1 个月前

Pango, for mcpp

Pango 1.56.1 as three mcpp packages, with upstream/ untouched and everything this fork adds under mcpp/.

[dependencies]
gnome.pangocairo = "1.56.1"   # the usual entry point: layout + rendering
# gnome.pango and gnome.pangoft2 arrive transitively — do NOT name them

member	what it is	module exports
`gnome.pango`	itemisation, bidi, line breaking, markup	851
`gnome.pangoft2`	fontconfig picks the file, FreeType rasterises	88
`gnome.pangocairo`	the cairo backend — `pango_cairo_show_layout`	284

Two ways to consume it, and you pick one:

import gnome.pangocairo;      // re-exports pango, pangoft2 and cairo

#include <pango/pangocairo.h>

They do not compose: a TU that does both reaches <stdio.h> twice — once through the module's global fragment, once directly — and the same struct _IO_FILE becomes two entities. Which route to take is decided by macros, which a module cannot carry: code using PANGO_TYPE_* or G_OBJECT takes the header route, code using the function API imports and includes nothing. Each member ships a test for each route, so both stay working.

Why a fork

Generators, not line count — the criterion that made cairo (104k lines) a plain descriptor and libdisplay-info (2k) a fork. Pango has three, and this fork adds a fourth:

upstream	here
`configure_file` → `pango-features.h`	`gen_features()`
`configure_file` → `config.h` (no input template)	`gen_config()`
`gobject/glib-mkenums` (816 lines of Python)	`write_enumtypes()`, 27 GTypes
(none — this fork's own)	`gen_module()` → three `.cppm`

There is no sh and no python in the build. build.mcpp is a compiled C++ program.

⚠️ config.h is the odd one: upstream writes configure_file(output: 'config.h', configuration: pango_conf) with no input:, so meson emits a #define per key and there is nothing in the tree to substitute into. The file has to be written, and every value in it is a decision this fork makes. Two of them are arithmetic rather than choice:

PANGO_BINARY_AGE    = minor * 100 + micro     # 5601
PANGO_INTERFACE_AGE = minor odd ? 0 : micro   # 1

pango_version_check() reads PANGO_BINARY_AGE, so a wrong value makes a correct version comparison answer wrongly at run time rather than failing to build. The test asserts both directions.

⭐ The one test that produces pixels

mcpp/pangocairo/tests/pangocairo.cpp renders "Hello 世界" to an ARGB32 surface and counts non-transparent pixels. Reaching that costs seven packages:

gnome.pango        itemisation, bidi, line breaking
gnome.pangoft2     fontconfig picks the file, FreeType rasterises
gnome.gio          PangoFontMap is a GListModel
compat.harfbuzz    shaping
compat.fribidi     the bidi algorithm
freedesktop.cairo  the surface the glyphs land on

A blank image means one of them is not doing its job.

⚠️ It degrades honestly when there are no fonts. freedesktop.fontconfig compiles its runtime paths empty on purpose, so a runner with no FONTCONFIG_FILE legitimately finds zero families — and then there is nothing to draw. That case is reported, not passed over, because "0 families, so the only real check did not run" and "the text rendered" must not look alike.

⚠️ A silent degradation this fork walked into

HAVE_CAIRO_FREETYPE was dropped from config.h by an editing slip. Nothing failed to build. pangocairo-fontmap.c simply registered no backends, and the program died at run time with

Pango-CRITICAL: Unknown $PANGOCAIRO_BACKEND value.
  Available backends are:            <- an empty list

followed by a segfault. The test now reads the font map's font type and requires CAIRO_FONT_TYPE_FT, which is the assertion that would have caught it immediately.

⚠️ `freedesktop.cairo`'s module is not sufficient on its own

Measured on 1.18.2: it exports 470 names, zero enumerators (no CAIRO_FORMAT_ARGB32, no CAIRO_FONT_TYPE_FT) and no cairo_t. The index's own cairo example does not notice, because it writes both #include <cairo.h> and import freedesktop.cairo; — the header supplies what the module lacks.

pangocairo cannot do that (mixing the routes is the struct _IO_FILE problem above), so gnome.pangocairo scans cairo.h itself and exports those names. Exporting the same entity from two modules is harmless — they are the same global-module entities — so this is additive. When cairo's wrapper is fixed, that scan can go.

⚠️ Name `pangocairo` alone

gnome.pango and gnome.pangoft2 are workspace path dependencies of gnome.pangocairo, and gnome.gio is a path dependency of gnome.pango. mcpp rejects a package requested both ways:

error: dependency 'gnome.gobject' is requested as both a version dep
       (by 'pango') and a path dep (by 'gnome.gio@2.82.5'). Pick one.

That error is how this fork learned it, on the first build.

Layout

upstream/              pango 1.56.1, byte for byte (CI diffs it)
mcpp/common/
  enums.h              glib-mkenums, ported from the glib fork
  prelude.h            gen_config / gen_features / gen_pango_enumtypes
  modules.h            gen_module, the .cppm wrappers
mcpp/pango/            + pango-features.h, pango-enum-types.{h,c}
mcpp/pangoft2/         + HAVE_FREETYPE
mcpp/pangocairo/       + HAVE_CAIRO, HAVE_CAIRO_FREETYPE

项目介绍

Pango 1.56.1 for mcpp — CN mirror

pango:Pango 1.56.1 for mcpp — CN mirror