BUG: ASV Benchmarks failing to build

[matteosantama]

ASV benchmarks fail to build on master. Could this be related to #34666?

error: command 'gcc' failed with exit status 1

Anyone else having this issue?

[TomAugspurger]

What command are you running to build them?

[matteosantama]

>>> git pull upstream master --ff-only
>>> asv continuous -f 1.1 upstream/master HEAD 

And then I get a bunch of errors like

1 error generated.
   pandas/_libs/tslibs/frequencies.c:7337:52: error: code will never be executed [-Werror,-Wunreachable-code]
           } else if (PY_VERSION_HEX >= 0x030700A0 && flag == (METH_FASTCALL | METH_KEYWORDS)) {
                                                      ^~~~
   pandas/_libs/tslibs/frequencies.c:7337:38: note: silence by adding parentheses to mark code as explicitly dead
           } else if (PY_VERSION_HEX >= 0x030700A0 && flag == (METH_FASTCALL | METH_KEYWORDS)) {
                                        ^
                                        /* DISABLES CODE */ ( )

My conda environment uses Python 3.8, which is why I thought maybe its connected to #34666

[jorisvandenbossche]

@matteosantama can you build pandas itself in your development environment (outside of the asv env)?

[matteosantama]

Yes, which makes it all the more confusing

[matteosantama]

It really doesn't like this line, which appears in 7 different files.

else if (PY_VERSION_HEX >= 0x030700A0 && flag == (METH_FASTCALL | METH_KEYWORDS))

My guess is that for some reason the condition PY_VERSION_HEX >= 0x030700A0 is failing, so I'm getting an unreachable code error on the second half of the statement. I'm also assuming that check confirms the python version is greater than 3.7. In the ASV benchmarks config file, line 29 looks like its setting the version to 3.6.

Lot of guesses here, so not sure if I'm even on the right track.

[jorisvandenbossche]

Can you check if you get the same problem or not when you change the asv config to use 3.7 or 3.8?

[jorisvandenbossche]

(that's a config that we should update at some point anyway)

[matteosantama]

Changing to 3.7 fixed it for me

[matteosantama]

I can make a PR if you think it's as simple as changing that line, unless you think there might be other ramifications you want to explore.

[jorisvandenbossche]

I think a change like that is welcome (but I would directly take 3.8, to not have to change this too often).

But, regardless of updating that, we still need to figure out why it is failing in a py3.6 environment, as we still support that.