package py37-tensorflow does not build on mac os X 10.15 (catalina)

[tiperiu]

The package py37-tensorflow does not build on Mac OS X 10.15, I installed xcode 11 and command-line tools and followed all the procedures.

I attached the main.log file. I did an update and upgrade outdated just before trying to install

thanks

[ryandesign (Ryan Carsten Schmidt)]

I see in the log:

ERROR: An error occurred during the fetch of repository 'nasm':

All mirrors are down: [GET returned 404 Not Found, connect timed out]

This demonstrates that the port has a bug: it tries to fetch something in the build phase. The port should be modified to fetch everything it needs in the fetch phase.

[cjones051073 (Chris Jones)]

Its not a bug in the port file, but a consequence of the fact tensorflow uses bazel to build, and bazel insists of doing this. I know of no way to avoid it.

[jxy (Xiao-Yong)]

The FreeBSD port,

​https://github.com/freebsd/freebsd-ports/blob/master/science/py-tensorflow/

though on 1.14.0, fetches everything in the fetch phase. It looks to me that you need to tell bazel to only fetch distfiles from the cache,

​https://github.com/freebsd/freebsd-ports/blob/master/science/py-tensorflow/files/bazelrc

and in the fetch phase fetch these to the cache,

​https://github.com/freebsd/freebsd-ports/blob/master/science/py-tensorflow/Makefile.MASTER_SITES

[cjones051073 (Chris Jones)]

thanks, that might be worth looking into but I suspect would not really help, because I don't see how we could make use of the fetched source in macports, for instance to add them to the mirrored distfiles. So every build would always need to fetch the source itself, in which case it really makes little difference which 'macports' phase it happens in. Its the same deal as with builds that insist on building from a full git checkout, you cannot really avoid every build having to fetch direct from the primary source...

[cjones051073 (Chris Jones)]

Looking into this, I sporadically see it, but just rerunning the build once or twice gets past it... Don't have a better solution at this to just 'keep trying til it works....'.

[jxy (Xiao-Yong)]

You can add the packages to the mirrored distfiles, or even a bundle of those, and copy them to the bazel cache during the extract phase.

Edit: to clarify, FreeBSD port does it in the post-patch phase.

[tiperiu]

I tried what @cjones051073 suggested and try sebveral time to insgtall the package, but now even a dependency py37-astor no longer builds. See attachment main2.log

[tiperiu]

this is the log from installing py37-astor (it fails)

[cjones051073 (Chris Jones)]

Please do not mix up issues with different ports in the same ticket. If you now have an issue with py-astor, you need to submit a new ticket for that.

[tiperiu]

I did another update and tried several times as @cjones051073 suggested but still no luck. No dependency problems, just the main package: py37-tensorflow

[mascguy (Christopher Nielsen)]

Support for Python 3.7/3.8 has been dropped for py-tensorflow, so this ticket is no longer relevant. Closing.