No fastcall attribute in POWER platform

[ayappanec]

Compiling the development branch of pandas fail in AIX.

pandas/_libs/src/ujson/lib/ultrajsonenc.c:397:1: error: ‘fastcall’ attribute directive ignored [-Werror=attributes]
Buffer_AppendShortHexUnchecked(char *outputOffset, unsigned short value) {
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
pandas/_libs/src/ujson/lib/ultrajsonenc.c:726:59: error: ‘fastcall’ attribute directive ignored [-Werror=attributes]
char *end) {
^~~~
cc1: all warnings being treated as errors

The fastcall attribute is not available in POWER platform also.

[WillAyd]

FWIW I would be OK with removing this altogether. AFAIK it's only for i386 platforms which we don't really care to optimize at this point

[ayappanec]

FWIW I would be OK with removing this altogether. AFAIK it's only for i386 platforms which we don't really care to optimize at this point

Okay, so shall i remove it altogether in this PR ?

[WillAyd]

Yea let's go ahead and remove it. If you can run the JSON benchmarks after than and post the results would help confirm

[ayappanec]

Not sure how to run those JSON benchmarks

[WillAyd]

Check this guide:

https://pandas.pydata.org/pandas-docs/stable/development/contributing.html#running-the-performance-test-suite

You can use the -b json switch at the end to limit the results to that module

[WillAyd]

If you can remove this altogether would be preferable. No need for an empty define

[ayappanec]

Removing the empty defines also needs lot of cleanup in ultrajsondec.c . Otherwise there will be compilation problems. And also not sure about the fastcall defines under _WIN32 .

[jreback]

looks fine @WillAyd merge when good

[WillAyd]

@ayappanec can you remove all occurrences of this macro instead of the empty define?

[ayappanec]

JSON_Benchmarks.txt
Attaching the JSON benchmarks results here

[jreback]

@ayappanec can you run vs the master commit, e.g. https://pandas.pydata.org/docs/dev/development/contributing.html#running-the-performance-test-suite, you need asv continuous

[ayappanec]

@ayappanec can you run vs the master commit, e.g. https://pandas.pydata.org/docs/dev/development/contributing.html#running-the-performance-test-suite, you need asv continuous

Looks like hdf5 needs to be installed . I never encountered hdf5 before

[ayappanec]

Installed HDF5 but now it fails with llvmlite build.

       raise RuntimeError("unsupported platform: %r" % (sys.platform,))
   RuntimeError: unsupported platform: 'aix6'

Looks like the asv benchmark testing needs some pre-requisites which has compilation issues in AIX .

[WillAyd]

lgtm

[ayappanec]

Any update on getting this merged ?

[jreback]

thanks @ayappanec

[ayappanec]

@WillAyd @jreback Thanks for the support.