1768116 - (C++20) [meta] Compile as C++20

Status: RESOLVED FIXED

(Firefox Build System :: General, task, P3)

Description

[Chris Peterson [:cpeterson]]

Who knows when we'll be able to compiled with -std=c++20, but in the meantime we can use this meta bug to gather the work needed to compile as C++20 and the new improvements that will become available with C++20.

When compiling as C++20, __cplusplus == 202002.

Summary of C++20 language changes: https://en.cppreference.com/w/cpp/20

Clang's C++20 support: https://clang.llvm.org/cxx_status.html#cxx20

GCC's C++20 support: https://gcc.gnu.org/projects/cxx-status.html#cxx20

Comment 1

[Chris Peterson [:cpeterson]]

This slide deck from a Google engineer details some of the problems (and fixes) found when updating Chrome from C++17 to C++20:

"C++20, How Hard Could It Be"
https://docs.google.com/presentation/d/1HwLNSyHxy203eptO9cbTmr7CH23sBGtTrfOmJf9n0ug/preview?resourcekey=0-GH5F3wdP7D4dmxvLdBaMvw

Chrome's C++20 meta bug:
https://bugs.chromium.org/p/chromium/issues/detail?id=1284275

Comment 2

[Kelsey Gilbert [:jgilbert]]

Thanks for working on this!

Comment 3

[Mike Hommey [:glandium]]

FYI, I don't expect we'd be able to actually switch to C++20 for a few years.

Comment 4

[Kelsey Gilbert [:jgilbert]]

What are our unlisted blockers? Our good friends hazard-builds and android toolchains?

FWIW, we went c++17 on 2019-12-03, so it's we're neither really ahead nor behind our previous pace.

Comment 5

[Chris Peterson [:cpeterson]]

(In reply to Kelsey Gilbert [:jgilbert] from comment #4)

What are our unlisted blockers? Our good friends hazard-builds and android toolchains?

IIUC, we still need to support clang 5.0 for some older Linux distros, but I don't know which distros are blocking us today. There is some 2019 discussion in bug 1579189 comment 6 about clang support on Debian 8 and 9, though Sylvestre replied that clang versions up to 8 were available on Debian 8 at that time.

Comment 6

[Corey Bryant]

Chris, Does Firefox have a time frame for switching to C++20? I ask because we have a library in Thunderbird that's gone EOL and the new version requires C++20.

Comment 7

[Chris Peterson [:cpeterson]]

(In reply to Corey Bryant from comment #6)

Does Firefox have a time frame for switching to C++20? I ask because we have a library in Thunderbird that's gone EOL and the new version requires C++20.

We don't have a definite ETA, but the priority has increased because Gecko depends on Google's Skia graphics and the next version will require C++20. The main blocker is updating our minimum clang, gcc, and libstdc++ versions in bug 1880779.

Just curious for future reference, what Thunderbird library will require C++20?

Comment 8

[Corey Bryant]

(In reply to Chris Peterson [:cpeterson] from comment #7)

(In reply to Corey Bryant from comment #6)

Does Firefox have a time frame for switching to C++20? I ask because we have a library in Thunderbird that's gone EOL and the new version requires C++20.

We don't have a definite ETA, but the priority has increased because Gecko depends on Google's Skia graphics and the next version will require C++20. The main blocker is updating our minimum clang, gcc, and libstdc++ versions in bug 1880779.

Just curious for future reference, what Thunderbird library will require C++20?

Chatted with Chris offline about this.

Comment 9

[Corey Bryant]

(In reply to Corey Bryant from comment #6)

Chris, Does Firefox have a time frame for switching to C++20? I ask because we have a library in Thunderbird that's gone EOL and the new version requires C++20.

Chatted with glandium and he said updating the minimum versions would come somewhere in the 140 or 141 cycle, probably.

Comment 10

[Chris Peterson [:cpeterson]]

Attachment: WIP: Bug 1768116 - Compile as C++20.

Comment 11

[Jeff Muizelaar [:jrmuizel]]

What's blocking this from landing?

Comment 12

[Chris Peterson [:cpeterson]]

(In reply to Jeff Muizelaar [:jrmuizel] from comment #11)

What's blocking this from landing?

After updating the base toolchain's minimum compiler versions, there are half a dozen new C++20 build errors that need to be fixed. They are mostly gcc issues that couldn't be tested with the older compiler versions. I have fixes for most of them. I can file bugs for each of them and submit the patches I have so far.

Comment 13

[Chris Peterson [:cpeterson]]

At this time, the remaining blockers to updating to C++20 (that I know of) are:

1. Bug 1992321: moz_xmalloc link errors on Linux. I don't know why bumping the GCC version and toolchain in bug 1981921 causes moz_xmalloc to become undefined when compiling as C++20 on both gcc and clang on Linux (but not with C++17 and not on macOS or Android). I could compile as C++20 with clang on Linux before bug 1981921.
2. Bug 1997382: objidl.idl build errors on Windows.
3. Update the test_toolchain_configure.py tests for C++20: https://searchfox.org/firefox-main/source/python/mozbuild/mozbuild/test/configure/test_toolchain_configure.py

Comment 14

[Chris Peterson [:cpeterson]]

Attachment: WIP: Bug 1768116 - Compile as C++20.

Comment 15

[Emilio Cobos Álvarez [:emilio]]

Attachment: Bug 1768116 - Compile as C++20. r=glandium!

Co-authored-by: Chris Peterson <[e-mail]>

Comment 16

[Emilio Cobos Álvarez [:emilio]]

Attachment: Bug 1768116 - updated libwebrtc moz.build files. r=mjf

Comment 17

[Emilio Cobos Álvarez [:emilio]]

Current try runs:

• Auto try run: https://treeherder.mozilla.org/jobs?repo=try&landoCommitID=163233
• All builds: https://treeherder.mozilla.org/jobs?repo=try&landoCommitID=163234

Comment 18

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/552f30ab626f
https://hg.mozilla.org/integration/autoland/rev/a076ddec6082
Compile as C++20. r=glandium,webrtc-reviewers,mjf
https://github.com/mozilla-firefox/firefox/commit/2fb967102611
https://hg.mozilla.org/integration/autoland/rev/47398d3d17dd
updated libwebrtc moz.build files. r=mjf

Comment 19

[Serban Stanca [:SerbanS]]

https://hg.mozilla.org/mozilla-central/rev/a076ddec6082
https://hg.mozilla.org/mozilla-central/rev/47398d3d17dd

Comment 20

[Florin Bilt]

(In reply to Serban Stanca [:SerbanS] from comment #19)

https://hg.mozilla.org/mozilla-central/rev/a076ddec6082
https://hg.mozilla.org/mozilla-central/rev/47398d3d17dd

Perfherder has detected a talos performance change from push 47398d3d17ddadd277fcdb1d60f24fdf61444dcd.

If you have any questions, please reach out to a performance sheriff. Alternatively, you can find help on Slack by joining #perf-help, and on Matrix you can find help by joining #perftest.

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
6%	perf_reftest_singletons input-type-number-setAttribute.html	linux1804-64-shippable-qr	e10s fission stylo webrender	191.19 -> 179.41

Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests.

If you need the profiling jobs you can trigger them yourself from treeherder job view or ask a performance sheriff to do that for you.

You can run all of these tests on try with ./mach try perf --alert 47419

The following documentation link provides more information about this command.