build: build v8 with -fvisibility=hidden -fvisibility-inlines-hidden

[joyeecheung]

Split from #56275 since it seems to cause gcc on some machines in the CI to time out or run out of memory. Trying to see if it's just a CI hiccup or if it's something that needs to be worked around.

[nodejs-github-bot]

Review requested:

• @nodejs/gyp
• @nodejs/v8-update

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/64084/

[richardlau]

FWIW changing compile/link options will invalidate any stored CCACHE entries, which would result in more of V8 having to be recompiled.

Generally most of the x64 CI machines in Jenkins are now 4 GB RAM + swap (2 GB).

Anything running on a machine with _container_ in its name will be sharing memory with other containers on the same host (albeit these hosts have a lot of memory (32 GB, IIRC)).

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/64102/

[joyeecheung]

In case anyone is curious, here is a bar chart I generated with ChatGPT about the busiest hours of the CI, using data from https://ci.nodejs.org/job/node-test-pull-request/api/json?tree=builds[timestamp,url,result] (past 100 jobs)

[bnoordhuis]

Dem 8 hour workday Europeans kicking off a quick CI run right before dinner, huh? That's a cool graph!

[joyeecheung]

(Unless you live in Spain, where 8PM is considered early for dinner).

In seriousness I think the spike is probably due to being an overlap of working hours between Europe and Americas, in any case 16-18 CET would be the time to avoid testing something that invalidates the ccache. I dumped the script to https://gist.github.com/joyeecheung/6006b7ee781a2a619021767e0c8f2c01 though I have no idea if it works locally.

Rebased and added -fvisibility-inlines-hidden. It seems the Node.js files are missing the flags too, though I am not yet sure if all the other headers are set up correctly for us to apply the flags directly without breaking addons, so I've commented it out.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/64429/

[joyeecheung]

Started a CI, though I noticed that test-orka-macos11-x64-1 had been offline for a couple of days, creating a very long osx build queue, I just cleaned it up and brought it online, so not sure how well it would fare after this CI gets its turn in the osx jobs. If it doesn't look great in the build load, then not adding -fvisibility-inlines-hidden on macOS for now would be a better idea otherwise it would make the osx build queue even longer

[joyeecheung]

Looks like the build is still timing out on some platforms, need to try again in a less busy hour...

There was an error on AIX that doesn't show up on other platforms https://ci.nodejs.org/job/node-test-commit-aix/55380/nodes=aix72-ppc64/testReport/junit/addons/heap-profiler/test/

duration_ms: 142.114
exitcode: 1
severity: fail
stack: "node:internal/modules/cjs/loader:1930\n  return process.dlopen(module, path.toNamespacedPath(filename));\n\
  \                 ^\n\nError: rtld: 0712-001 Symbol _ZN2v812OutputStream12GetChunkSizeEv\
  \ was referenced\n      from module /home/iojs/build/workspace/node-test-commit-aix/nodes/aix72-ppc64/test/addons/heap-profiler/build/Release/binding.node(),\
  \ but a runtime definition\n\t    of the symbol was not found.\nrtld: 0712-001 Symbol\
  \ _ZN2v812OutputStream19WriteHeapStatsChunkEPNS_15HeapStatsUpdateEi was referenced\n\
  \      from module /home/iojs/build/workspace/node-test-commit-aix/nodes/aix72-ppc64/test/addons/heap-profiler/build/Release/binding.node(),\
  \ but a runtime definition\n\t    of the symbol was not found.\n    at Object..node\
  \ (node:internal/modules/cjs/loader:1930:18)\n    at Module.load (node:internal/modules/cjs/loader:1473:32)\n\
  \    at Function._load (node:internal/modules/cjs/loader:1285:12)\n    at TracingChannel.traceSync\
  \ (node:diagnostics_channel:322:14)\n    at wrapModuleLoad (node:internal/modules/cjs/loader:234:24)\n\
  \    at Module.require (node:internal/modules/cjs/loader:1495:12)\n    at require\
  \ (node:internal/modules/helpers:135:16)\n    at Object.<anonymous> (/home/iojs/build/workspace/node-test-commit-aix/nodes/aix72-ppc64/test/addons/heap-profiler/test.js:5:17)\n\
  \    at Module._compile (node:internal/modules/cjs/loader:1739:14)\n    at Object..js\
  \ (node:internal/modules/cjs/loader:1904:10) {\n  code: 'ERR_DLOPEN_FAILED'\n}\n\
  \nNode.js v24.0.0-pre"

AFAICT, it shouldn't be failing as v8::OutputStream is marked as V8_EXPORT and it works fine on other platforms (the test also seemed to work fine from the CI started 3 weeks ago). Pinging @nodejs/platform-aix to see what's going on, if there's no one who cares about binary size on AIX to fix it I will just skip the settings on AIX I guess.

[richardlau]

There was an error on AIX that doesn't show up on other platforms https://ci.nodejs.org/job/node-test-commit-aix/55380/nodes=aix72-ppc64/testReport/junit/addons/heap-profiler/test/

AFAICT, it shouldn't be failing as v8::OutputStream is marked as V8_EXPORT and it works fine on other platforms (the test also seemed to work fine from the CI started 3 weeks ago). Pinging @nodejs/platform-aix to see what's going on, if there's no one who cares about binary size on AIX to fix it I will just skip the settings on AIX I guess.

Seems strange that it worked 3 weeks ago but not in the recent run.

Symbol exporting is different on AIX -- we have https://github.com/nodejs/node/blob/main/tools/create_expfile.sh to generate an exports file containing the symbols that is used for addons:

node/tools/create_expfile.sh

Lines 2 to 6 in 529b56e

	# Writes out all of the exported symbols to a file.
	# This is needed on AIX as symbols are not exported
	# by an executable by default and need to be listed
	# specifically for export so that they can be used
	# by native add-ons.

We haven't touched it in five years -- I'm not even sure it's needed anymore since https://gcc.gnu.org/gcc-7/changes.html has

Visibility support has been enabled for AIX 7.1 and above.

@abmusse Is this something you can look into?

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/64810/

[richardlau]

There was an error on AIX that doesn't show up on other platforms https://ci.nodejs.org/job/node-test-commit-aix/55380/nodes=aix72-ppc64/testReport/junit/addons/heap-profiler/test/

AFAICT, it shouldn't be failing as v8::OutputStream is marked as V8_EXPORT and it works fine on other platforms (the test also seemed to work fine from the CI started 3 weeks ago). Pinging @nodejs/platform-aix to see what's going on, if there's no one who cares about binary size on AIX to fix it I will just skip the settings on AIX I guess.

@joyeecheung Since this still occurs (https://ci.nodejs.org/job/node-test-commit-aix/55692/nodes=aix72-ppc64/testReport/junit/addons/heap-profiler/test/) please skip the setting on AIX. We (IBM/RH) will add it to the list of things to look at on AIX.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/69760/

[joyeecheung]

Rebased this after I realized how big the binary has become on Linux recently. This currently gives

ubuntu@ip-172-31-7-86:~/node$ ls -lah out/Release/node
-rwxrwxr-x 1 ubuntu ubuntu 120M Oct 16 15:26 out/Release/node
ubuntu@ip-172-31-7-86:~/node$ ls -lah ./node_main
-rwxrwxr-x 1 ubuntu ubuntu 128M Oct 16 14:19 ./node_main

Trying to see if CI cache fares better with the flag changes now..

[legendecas]

Can this be moved up above the V8_TLS_USED_IN_LIBRARY? So that this can be closer to the comments above, which is less relevant with V8_TLS_USED_IN_LIBRARY.

[joyeecheung]

Would you mind if I leave it to a follow-up PR? Trying to get the cache of it into the CI first instead of retrying for flakes and make the cache a mess in the CI...

[joyeecheung]

Merged it to put the cache in, feel free to open another PR to move them closer

[nodejs-github-bot]

Landed in 3437e1c