GCC 14.1, 13.3 & 12.4 #133

h-vetinari · 2024-05-07T22:13:26Z

Freshly released: 14.1, 13.3, 12.4

conda-forge-webservices · 2024-05-07T22:13:34Z

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

h-vetinari · 2024-05-07T23:46:45Z

@conda-forge/ctng-compilers, this is ready for review! Point releases for the maintenance branches should show up towards the end of May.

h-vetinari · 2024-05-10T03:12:32Z

Ah, we might want to fix #126 here already.

h-vetinari · 2024-05-13T21:54:08Z

@conda-forge-admin, please rerender

h-vetinari · 2024-05-13T22:04:58Z

Not sure what's up after #130, but I cannot rerender locally anymore.

  File "E:\miniforge\envs\builder\Lib\site-packages\conda_smithy\feedstock_io.py", line 55, in write_file
    with io.open(filename, "w", encoding="utf-8", newline="\n") as fh:
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
FileNotFoundError: [Errno 2] No such file or directory: 'C:\\Users\\[xxx]\\dev\\conda-forge\\ctng-compilers-feedstock\\.ci_support\\linux_aarch64_cross_target_platformlinux-64cross_target_stdlibsysrootcross_target_stdlib_version2.12gcc_maj_ver11gcc_version11.4.0old_tripletx86_64-conda_cos6-linux-gnutripletx86_64-conda-linux-gnu.yaml'

Might be running into path length limitations on windows...

At least the bot still can. :)

…nda-forge-pinning 2024.05.13.13.50.55

h-vetinari · 2024-05-13T22:50:14Z

@isuruf, unsurprisingly, the mingw patches from #130 fail against GCC 14. What would you like to do here - carry two sets of patches, or build mingw only against GCC 13 (or only against 14)?

isuruf · 2024-05-13T22:56:24Z

We can carry two sets in recipe/patches/mingw/13 and recipe/patches/mingw/14

h-vetinari · 2024-05-14T19:55:03Z

@conda-forge/ctng-compilers, this should be ready for review. We might want to link the C++ chrono implementation to a tzdb we supply ourselves (with tzdata; see #126), but for now this is a vanilla upgrade.

h-vetinari · 2024-05-21T13:34:32Z

@conda-forge-admin, please rerender

…nda-forge-pinning 2024.05.21.12.55.42

h-vetinari · 2024-05-21T20:52:20Z

This now also contains 13.3 & the tzdb fix. PTAL @conda-forge/ctng-compilers :)

isuruf · 2024-05-22T02:24:56Z

Thanks. 13.3 build fails to apply the patches though.

isuruf · 2024-05-22T02:56:51Z

For tzdb, can you add a test? Binary prefix relocation usually don't work when the prefix is stored in a std::string.

h-vetinari · 2024-06-17T22:21:38Z

OK, if you prefer platform-specific implementations over environment variables, that's fine by me. 🤷

h-vetinari · 2024-06-20T07:11:22Z

OK, if you prefer platform-specific implementations over environment variables, that's fine by me. 🤷

So I tried looking at this, and it's not trivial (from what I can see) to get the path to the library where the function lives. It'd be a bit easier to get the path of the calling executable, but that's useless because it has unknown relationship to $PREFIX. C++20's std::source_location also doesn't help. The best-looking solution I found is the small whereami library (particularly wai_getModulePath), but copying/vendoring that seems like major overkill for something that's solved in a handful of lines when using environment variables...

Or could you tell me how you you envisioned getting a path at a stable/predictable depth in the environment file hierarchy from the C++ side?

isuruf · 2024-06-20T17:03:36Z

See https://github.com/gcc-mirror/gcc/blob/9a76db24e044c8058497051a652cca4228cbc8e9/libsanitizer/sanitizer_common/sanitizer_dl.cpp#L25-L35 for an example.

h-vetinari · 2024-06-20T23:41:15Z

See https://github.com/gcc-mirror/gcc/blob/9a76db24e044c8058497051a652cca4228cbc8e9/libsanitizer/sanitizer_common/sanitizer_dl.cpp#L25-L35 for an example.

Thanks. I found the dladdr approach too, but obviously that won't work for windows, where we'd have to use GetModuleFileNameW or similar.

…nda-forge-pinning 2024.06.20.18.47.46

h-vetinari · 2024-06-21T23:28:38Z

So I implemented something for the relative path approach here, but obviously using dladdr on linux means we need to link libstd++ to libdl. For now I've just hacked this with LDFLAGS (and even that I cannot get to work once something links to libstd++.so itself), but what I'd like to do is use the equivalent of CMakes target_link_libraries for libstd++.so itself.

I think it should be something like AC_LIB_LINKFLAGS, but the makefile(s) in GCC are just indecipherable for me (though I really tried to find where the shared lib gets defined...). Any ideas how/where to do this?

mbargull · 2024-06-22T07:09:25Z

recipe/patches/0004-patch-zoneinfo_dir_override-to-point-to-our-tzdata.patch

+        tz_dir += "/share/zoneinfo";
+        // std::string constructor deep-copies
+        free(this_lib);
+        return tz_dir.c_str();


This pointer outlives the memory it points to when it is being returned.
We should return a pointer to statically allocated memory instead, I think.
Additionally, the allocations from this_lib and tz_dir are not really needed.
For this_lib you can just continue to use resolved instead.
And if you allocate buffer with size PATH_MAX + sizeof(TO_PREFIX "/share/zoneinfo") you can append the suffix directly to resolved in buffer's memory then.

Thanks! For now, I'm just trying to get something working¹. Improving the implementation can be done afterwards (even if it happens as part of this PR after all).

Footnotes

despite trying to keep the implementation minimal/simple, I remain strongly doubtful that the trade-off of all that complexity just for avoiding a corner³ case compared to the env var approach is beneficial. ↩

Improving the implementation can be done afterwards (even if it happens as part of this PR after all).

This was just me seeing a use-after-free error, so wanted to make sure that we don't include it in lower level libs.

I remain strongly doubtful that the trade-off of all that complexity just for avoiding a corner case compared to the env var approach is beneficial.

<chrono> use and in conjunction with hitting tzdb absence/differences is a corner case as of now, yes.
Not having CONDA_PREFIX set or set to a different path is really common, though (e.g., think about conda-build: we have a host and build environment -- only packages from one of them can use CONDA_PREFIX at runtime).
There are plethora of ways things can go wrong with reliance on env vars (and with reliance on activation scripts even more so) -- it's too brittle for us to rely on in lower level libs.
(Wcan chat about why/where there are so many problems with that if things are unclear, if you want.)

Not having CONDA_PREFIX set or set to a different path is really common, though (e.g., think about conda-build: we have a host and build environment -- only packages from one of them can use CONDA_PREFIX at runtime).

It's easy to probe all of CONDA_PREFIX resp. PREFIX / BUILD_PREFIX (if CONDA_BUILD is set), and even without anything, we'd find the system tzdb.

So a bug related to this requires: no valid env vars (which is not the case in any of our build infra or user envs), an out-of-date system tzdb (<-- user's fault/problem IMO), stumbling over anything that actually changed in the tzdb compared to an older system tzdb (almost nothing changes unless a country changes its laws about timezones), and using a timestamp just around the summer-/wintertime change. It's so extremely narrow a problem that I don't think the implementation effort for relative paths is anywhere near justified.

In any case, I'm trying to follow your and Isuru's preference. Help with the implementation is appreciated (so thanks for pointing out the bug) - I didn't expect it to be ready yet either, hence the debugging skips here. Right now I'm not even hitting the UAF due to the problems related to linking libdl, so that's the first thing to tackle.

h-vetinari · 2024-06-23T21:04:11Z

I'm going to go with this for now:

@isuruf: Let's remove the patch and merge and figure out tzdb stuff later.

I'll put up another PR with the WIP patch, and we can work from there, without indefinitely blocking the version upgrade.

h-vetinari · 2024-06-24T19:41:13Z

Any objections to merging @isuruf?

h-vetinari · 2024-06-27T22:57:49Z

recipe/meta.yaml

@@ -30,7 +30,7 @@ package:
  version: {{ version }}

 source:
-  - url: https://sourceware.org/pub/gcc/releases/gcc-{{ version }}/gcc-{{ version }}.tar.gz
+  - url: https://ftp.gnu.org/gnu/gcc/gcc-{{ version }}/gcc-{{ version }}.tar.gz


FWIW, 12.4 wasn't available on the ftp server shortly after release, and the announcement made it seem like that won't happen (also my understanding of sourceware was that it's much closer to GCC than third-party). Just for context why I changed that.

Ah, thanks. I didn't see the FSF affiliation.

h-vetinari · 2024-06-28T04:24:16Z

@isuruf, AFAICT we have a bootstrapping problem here:

Could not solve for environment specs
The following package could not be installed
└─ gcc_impl_win-64 13.3.0.*  is not installable because it requires
   └─ libgcc-devel_win-64 13.3.0 h5200ebd_100, which requires
      └─ __win, which is missing on the system.

libgcc-devel_win-64 is being built in that same job, but gcc (the output where things fail) wasn't part of the first mingw builds, so by the time it was enabled (e60f041) we already had some libgcc builds to work with.

I guess the subdir confusion is coming from a missing build: environment (even if it's empty, like it is for gcc_impl).

h-vetinari requested review from beckermr, isuruf, timsnyder and xhochy as code owners May 7, 2024 22:13

h-vetinari added 4 commits May 14, 2024 08:21

add GCC 14.1, drop 11.4

3de056e

renormalize patches

b26f072

rebase patches

13776b0

bump build number

c25ac86

h-vetinari force-pushed the 14 branch from 3a18a5a to c25ac86 Compare May 13, 2024 21:53

MNT: Re-rendered with conda-build 24.3.0, conda-smithy 3.35.1, and co…

8722e15

…nda-forge-pinning 2024.05.13.13.50.55

h-vetinari added 3 commits May 14, 2024 17:21

move mingw patches into version-specific folder

b68fb43

add mingw patches for GCC 14

d6b8ffb

fix patch names between gcc 13/14

eaf6ac1

h-vetinari added 3 commits May 21, 2024 12:10

set --with-libstdcxx-zoneinfo

0377ad8

let gxx_impl depend on tzdata

3e2d737

GCC 13.3.0

3a3aeac

h-vetinari changed the title ~~GCC 14.1~~ GCC 14.1 & 13.3 May 21, 2024

MNT: Re-rendered with conda-build 24.5.0, conda-smithy 3.36.0, and co…

5cd75eb

…nda-forge-pinning 2024.05.21.12.55.42

h-vetinari changed the title ~~GCC 14.1 & 13.3~~ GCC 14.1, 13.3 & 12.4 Jun 20, 2024

gcc 12.4.0

2662cc7

h-vetinari added 2 commits June 21, 2024 10:47

MNT: Re-rendered with conda-build 24.5.1, conda-smithy 3.36.2, and co…

01a38cf

…nda-forge-pinning 2024.06.20.18.47.46

Merge remote-tracking branch 'upstream/main' into 14

c857bdc

h-vetinari force-pushed the 14 branch from 0d603b2 to ffea9d2 Compare June 21, 2024 02:59

h-vetinari mentioned this pull request Jun 22, 2024

libcxx v18.1.8 conda-forge/libcxx-feedstock#155

Merged

8 tasks

mbargull reviewed Jun 22, 2024

View reviewed changes

h-vetinari force-pushed the 14 branch from 45def25 to c517e7c Compare June 23, 2024 21:03

h-vetinari force-pushed the 14 branch from c517e7c to 98e1155 Compare June 23, 2024 21:05

h-vetinari added 2 commits June 24, 2024 08:14

Merge remote-tracking branch 'upstream/main' into 14

32e291a

remove tzdb integration

2c40af6

h-vetinari force-pushed the 14 branch from 98e1155 to 2c40af6 Compare June 23, 2024 21:14

Use upstream URL instead of third party

7099a49

h-vetinari commented Jun 27, 2024

View reviewed changes

isuruf merged commit da577d5 into conda-forge:main Jun 28, 2024
19 of 64 checks passed

h-vetinari deleted the 14 branch June 28, 2024 00:55

This was referenced Jun 28, 2024

GCC 14.1, 13.3, 12.4 conda-forge/ctng-compiler-activation-feedstock#121

Merged

gfortran 14.1, 13.3, 12.4 conda-forge/gfortran_impl_osx-64-feedstock#82

Open

tzdb integration #142

Merged

This was referenced Jun 28, 2024

add empty build environment to gcc to break subdir confusion #143

Closed

install conda-build using pip #144

Merged

GCC 13 & LLVM 17 conda-forge/conda-forge-pinning-feedstock#4890

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

GCC 14.1, 13.3 & 12.4 #133

GCC 14.1, 13.3 & 12.4 #133

h-vetinari commented May 7, 2024 •

edited

Loading

conda-forge-webservices bot commented May 7, 2024

h-vetinari commented May 7, 2024

h-vetinari commented May 10, 2024

h-vetinari commented May 13, 2024

h-vetinari commented May 13, 2024

h-vetinari commented May 13, 2024

isuruf commented May 13, 2024

h-vetinari commented May 14, 2024

h-vetinari commented May 21, 2024

h-vetinari commented May 21, 2024

isuruf commented May 22, 2024

isuruf commented May 22, 2024

h-vetinari commented Jun 17, 2024

h-vetinari commented Jun 20, 2024

isuruf commented Jun 20, 2024

h-vetinari commented Jun 20, 2024

h-vetinari commented Jun 21, 2024

mbargull Jun 22, 2024 •

edited

Loading

h-vetinari Jun 22, 2024

mbargull Jun 22, 2024

h-vetinari Jun 22, 2024

h-vetinari commented Jun 23, 2024

h-vetinari commented Jun 24, 2024

h-vetinari Jun 27, 2024

isuruf Jun 27, 2024

h-vetinari commented Jun 28, 2024

GCC 14.1, 13.3 & 12.4 #133

GCC 14.1, 13.3 & 12.4 #133

Conversation

h-vetinari commented May 7, 2024 • edited Loading

conda-forge-webservices bot commented May 7, 2024

h-vetinari commented May 7, 2024

h-vetinari commented May 10, 2024

h-vetinari commented May 13, 2024

h-vetinari commented May 13, 2024

h-vetinari commented May 13, 2024

isuruf commented May 13, 2024

h-vetinari commented May 14, 2024

h-vetinari commented May 21, 2024

h-vetinari commented May 21, 2024

isuruf commented May 22, 2024

isuruf commented May 22, 2024

h-vetinari commented Jun 17, 2024

h-vetinari commented Jun 20, 2024

isuruf commented Jun 20, 2024

h-vetinari commented Jun 20, 2024

h-vetinari commented Jun 21, 2024

mbargull Jun 22, 2024 • edited Loading

Choose a reason for hiding this comment

h-vetinari Jun 22, 2024

Choose a reason for hiding this comment

Footnotes

mbargull Jun 22, 2024

Choose a reason for hiding this comment

h-vetinari Jun 22, 2024

Choose a reason for hiding this comment

h-vetinari commented Jun 23, 2024

h-vetinari commented Jun 24, 2024

h-vetinari Jun 27, 2024

Choose a reason for hiding this comment

isuruf Jun 27, 2024

Choose a reason for hiding this comment

h-vetinari commented Jun 28, 2024

h-vetinari commented May 7, 2024 •

edited

Loading

mbargull Jun 22, 2024 •

edited

Loading