build: initial support for meson build system #902

mochaaP · 2025-02-08T10:53:25Z

Initial support for meson build system.
I plan to maintain this port continuously alongside cmake.

suppoted features

~~make BUILD_BACKEND=meson~~ removed per request on build: initial support for meson build system #902 (comment)
all current cmake options (except wasi builds)
tests integration
benchmarks integration

xrefs

mesonbuild/wrapdb#1865
#761

tested

meson.build

dcbaker · 2025-02-08T17:57:15Z

meson.build

+  generator(
+    find_program('cp', 'xcopy'),
+    output: ['@PLAINNAME@'],
+    arguments: ['@INPUT@', '@OUTPUT@'],
+  ).process(qjs_hdrs),


There is import('fs').copyfile() for this. Though I'm confused why you're copying the header at all. Although it seems like you're doing this to try to put the header into a unique directory?

generator puts this into the target's private directory, and i reused that for declare_dependency to expose only the public headers.

i agree this is dirty. it should be better to change the directory structure, but @saghul might want to comment on this first.

Changing the directory structure is not on the table, sorry.

Many users still embed QuickJS by directly referencing the handful of files and we'd like to keep it that way for now.

Understandable.

The other workaroud people will use for this it to have an include/meson.build that does the copying so that it gets laid out correctly in the working directory. This is a long standing but thorny issue for us (Meson) to fix.

meson.build

examples/meson.build

dcbaker · 2025-02-08T18:08:32Z

meson.build

+qjsc_exe = executable(
+  'qjsc',
+  qjsc_srcs,
+  qjs_libc_srcs,


This will cause all of these sources to be re-compiled for the executable, but then it seems like you link them back in via the qjs_dep.

qjs_libc_srcs == [] if -Dlibc=true, since it's already compiled in the main qjs_lib.
this variable is here to explicitly add quickjs-libc to required targets. is it better to use qjs_libc_lib = static_library(...) and then add link_whole: qjs_libc_lib to qjs_lib?

applied the qjs_libc_lib thing.

dcbaker · 2025-02-08T18:10:03Z

I saw your comments on the wrap proposal, and thought I might take a peak at what you where working on.

saghul · 2025-02-08T19:03:35Z

Prior art: #761

trim21 · 2025-02-08T20:10:22Z

I'm the author of mesonbuild/wrapdb#1865 . Then initial motivation is that we do not want to add meson.build in this project. If we are going to support meson build system natively I'll close mesonbuild/wrapdb#1865 because it's unnecessary

saghul · 2025-02-08T20:49:49Z

I'm the author of mesonbuild/wrapdb#1865 . Then initial motivation is that we do not want to add meson.build in this project. If we are going to support meson build system natively I'll close mesonbuild/wrapdb#1865 because it's unnecessary

Personally I'm still on the fence about adding Meson support here. If wrapdb is there then users already have a good way forward.

bnoordhuis · 2025-02-08T21:45:36Z

I'm mildly negative on the changes to the Makefile. A meson build file, okay, but only if I don't have to look at it or think about it too much. For testing it, I'd add a CI job.

dcbaker · 2025-02-08T21:47:01Z

From Meson's perspective we would prefer if meson.build files could live upstream, even with a lower level of support that $main_build_system. WrapDB can still reference projects with upstream meson.build files to make meson wrap install quickjs-ng work seemlessly.

meson.build

meson_options.txt

dcbaker · 2025-02-08T23:19:10Z

LGTM from a meson is correct point of view

saghul · 2025-02-09T07:38:55Z

@bnoordhuis whats your appetite towards this? I'm somewhere around -0.5

chqrlie · 2025-02-09T08:21:59Z

I'm mildly negative on the changes to the Makefile. A meson build file, okay, but only if I don't have to look at it or think about it too much. For testing it, I'd add a CI job.

Regarding changes to the Makefile and a choice of a build option, I am nor negative but I would favor an option to build directly from Makefile on selected targets, without CMake :)

saghul · 2025-02-09T08:55:17Z

@chqrlie Please let's keep this about meson.

bnoordhuis · 2025-02-09T14:48:24Z

@bnoordhuis whats your appetite towards this? I'm somewhere around -0.5

I'm okay with merging it provided it's close-to-zero effort to maintain. I don't care for it myself but there's no denying meson is pretty popular nowadays.

Having said that, since wrapdb exists, is there a compelling reason to have the build file in this repo?

dcbaker · 2025-02-09T18:50:17Z

Having them upstream means that your users can use git snapshots and git submodules as subprojects, that’s not possible with meson files hosted in wrapdb

bnoordhuis · 2025-02-09T21:06:44Z

That's a good reason. @saghul, up to you.

saghul · 2025-02-10T09:00:55Z

Alright, let's give it a try. I'll give it a deeper look this week. What this PR is missing is a few CI jobs, to make sure this thing works at all :-)

mochaaP · 2025-02-10T10:22:09Z

sure, i will add it

mochaaP · 2025-02-10T10:30:20Z

should I add this to a separate workflow or just ci.yml?

saghul · 2025-02-10T11:09:49Z

Same file, new jobs.

mochaaP · 2025-02-10T12:39:38Z

clang sanitized builds fail with: runtime error: call to function js_realloc_rt through pointer to incorrect function type 'void *(*)(void *, void *, unsigned long)

saghul · 2025-02-10T17:30:09Z

In quickjs-libc?

Any way to get more output from the meson runners? They seem to output to a file.

mochaaP · 2025-02-10T17:58:10Z

@saghul https://nightly.link/quickjs-ng/quickjs/actions/runs/13241042765

saghul · 2025-02-10T19:35:34Z

Can you try if this helps?

diff --git a/quickjs-libc.c b/quickjs-libc.c
index e784878..c8892d2 100644
--- a/quickjs-libc.c
+++ b/quickjs-libc.c
@@ -170,9 +170,15 @@ typedef struct JSThreadState {
 
 static uint64_t os_pending_signals;
 
+static void *js_std_dbuf_realloc(void *opaque, void *ptr, size_t size)
+{
+    JSRuntime *rt = opaque;
+    return js_realloc_rt(rt, ptr, size);
+}
+
 static void js_std_dbuf_init(JSContext *ctx, DynBuf *s)
 {
-    dbuf_init2(s, JS_GetRuntime(ctx), (DynBufReallocFunc *)js_realloc_rt);
+    dbuf_init2(s, JS_GetRuntime(ctx), js_std_dbuf_realloc);
 }
 
 static bool my_isdigit(int c)

saghul · 2025-02-11T08:18:16Z

Looks like tests are failing here now: Failed: JS_IsJobPending(rt), file ../interrupt-test.c, line 114 but I have no idea why 😮

bnoordhuis · 2025-02-13T08:58:51Z

Can meson be configured to just print the test output instead of logging it to a file? I don't want to download and unzip a file every time something goes wrong.

The test failure suggests somehow the async function job didn't get scheduled, although I have no idea how that could happen.

mochaaP · 2025-02-13T12:38:01Z

@bnoordhuis addressed in c354670

saghul · 2025-02-13T12:40:08Z

I wonder if we simply don't run that test under any sanitizer in the cmake build.

bnoordhuis · 2025-02-13T14:03:19Z

It's tested under both asan+ubsan and msan and passes successfully.

bnoordhuis · 2025-02-13T21:44:27Z

@mochaaP does the test fail or pass locally for you?

mochaaP · 2025-02-15T01:19:08Z

yes, it fails:

saghul · 2025-02-15T06:52:08Z

Can you spots what do the failed jobs have in common?

bnoordhuis · 2025-02-15T19:24:07Z

.github/workflows/ci.yml

+      # ASLR with big PIE slides does not work well with [AM]San
+      - name: disable ASLR
+        if: ${{ matrix.platform == 'ubuntu-latest' && contains(matrix.mode.name, 'san') }}
+        run: |
+          sudo sysctl -w kernel.randomize_va_space=0
+


Suggested change

# ASLR with big PIE slides does not work well with [AM]San

- name: disable ASLR

if: ${{ matrix.platform == 'ubuntu-latest' && contains(matrix.mode.name, 'san') }}

run: |

sudo sysctl -w kernel.randomize_va_space=0

Should not be necessary anymore after 9c25179

bnoordhuis · 2025-02-15T19:28:51Z

meson.build

+  # Interrupt test
+  test(
+    'interrupt',
+    executable(
+      'interrupt-test',
+      'interrupt-test.c',


Was renamed in 22cd6ab.

Suggested change

# Interrupt test

test(

'interrupt',

executable(

'interrupt-test',

'interrupt-test.c',

# API test

test(

'api',

executable(

'api-test',

'api-test.c',

bnoordhuis · 2025-02-15T19:29:59Z

meson.build

+      'bigint64_arith',
+      'bigint256_arith',
+    ]
+      benchmark(


When does this run? If the CI doesn't exercise it, it's virtually guaranteed to get out of sync sooner or later.

…suppress clang sanitize warnings Co-authored-by: Saúl Ibarra Corretgé <[email protected]>

mochaaP · 2025-02-19T03:41:53Z

@saghul:

Can you spots what do the failed jobs have in common?

After a few days of troubleshooting, seems the tests will only fail with -O0 on clang. Should we use debugoptimized instead?

mochaaP force-pushed the mochaa branch from cbd715f to a5896a6 Compare February 8, 2025 13:30

dcbaker reviewed Feb 8, 2025

View reviewed changes

mochaaP force-pushed the mochaa branch 2 times, most recently from 24aba70 to 42cb912 Compare February 8, 2025 20:09

mochaaP force-pushed the mochaa branch from 42cb912 to 2d5f376 Compare February 8, 2025 20:12

trim21 mentioned this pull request Feb 8, 2025

quickjs-ng: new wrap mesonbuild/wrapdb#1865

Closed

dcbaker reviewed Feb 8, 2025

View reviewed changes

meson.build Show resolved Hide resolved

dcbaker reviewed Feb 8, 2025

View reviewed changes

meson.build Outdated Show resolved Hide resolved

dcbaker reviewed Feb 8, 2025

View reviewed changes

meson_options.txt Outdated Show resolved Hide resolved

mochaaP force-pushed the mochaa branch 2 times, most recently from 3c67859 to d5deb43 Compare February 10, 2025 11:41

bnoordhuis reviewed Feb 15, 2025

View reviewed changes

mochaaP and others added 15 commits February 16, 2025 22:14

build: initial support for meson build system

e283bbe

build: compile native qjs cli for subproject tool usage

caa2989

fixup! build: initial support for meson build system

c49851d

fixup! build: initial support for meson build system

29ee5b1

build: revert makefile changes

3ee2b1a

ci: add jobs for meson build

782f069

ci: fix actionlint warnings

2be3004

fixup! ci: add jobs for meson build

a1f95d1

fixup! ci: add jobs for meson build

5652495

fixup! ci: add jobs for meson build

67922b1

quickjs-libc: cast opaque pointer to JSRuntime* in js_std_dbuf_init, …

bbc73b3

…suppress clang sanitize warnings Co-authored-by: Saúl Ibarra Corretgé <[email protected]>

fixup! ci: add jobs for meson build

d6389dd

build: fix test_fib not correctly exporting qjs symbols

5c144b1

fixup! build: initial support for meson build system

033fb95

fixup! ci: add jobs for meson build

3fc17e4

mochaaP force-pushed the mochaa branch from a0a0313 to 3fc17e4 Compare February 16, 2025 14:17

fixup! ci: add jobs for meson build

438fd84

mochaaP force-pushed the mochaa branch from 792ef77 to 438fd84 Compare February 19, 2025 04:15

build: initial support for meson build system #902

Are you sure you want to change the base?

build: initial support for meson build system #902

Conversation

mochaaP commented Feb 8, 2025 • edited Loading

suppoted features

xrefs

tested

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dcbaker commented Feb 8, 2025

saghul commented Feb 8, 2025

trim21 commented Feb 8, 2025

saghul commented Feb 8, 2025

bnoordhuis commented Feb 8, 2025

dcbaker commented Feb 8, 2025

dcbaker commented Feb 8, 2025

saghul commented Feb 9, 2025

chqrlie commented Feb 9, 2025

saghul commented Feb 9, 2025

bnoordhuis commented Feb 9, 2025

dcbaker commented Feb 9, 2025

bnoordhuis commented Feb 9, 2025

saghul commented Feb 10, 2025

mochaaP commented Feb 10, 2025

mochaaP commented Feb 10, 2025

saghul commented Feb 10, 2025

mochaaP commented Feb 10, 2025 • edited Loading

saghul commented Feb 10, 2025

mochaaP commented Feb 10, 2025

saghul commented Feb 10, 2025

saghul commented Feb 11, 2025

bnoordhuis commented Feb 13, 2025

mochaaP commented Feb 13, 2025

saghul commented Feb 13, 2025

bnoordhuis commented Feb 13, 2025

bnoordhuis commented Feb 13, 2025

mochaaP commented Feb 15, 2025

saghul commented Feb 15, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mochaaP commented Feb 19, 2025

mochaaP commented Feb 8, 2025 •

edited

Loading

mochaaP commented Feb 10, 2025 •

edited

Loading