[Python-Dev] #ifdeffery (original) (raw)

Michael Hudson mwh at python.net
Fri Aug 13 13:05:20 CEST 2004

"Raymond Hettinger" <python at rcn.com> writes:

P.S. Side rant: Also, in the last few months, the #ifdefs have multiplied. While I understand that some things like time stamping are worthwhile, the code is getting harder to read and maintain because of all this b.s.

Is it really getting worse?

I think there are two qualitatively different sorts of #ifdef mess: feature ones, like the WITH_TSC I've recently been fiddling with recently and portability ones.

The feature #ifdefs aren't (or shouldn't be) that much of a big deal. Ideally, they are documented in Misc/SpecialBuilds.txt and depending on whether they are defined or not, bits and pieces of code are or are not included.

Platform #ifdefs, on the other hand, are pain in a tin.

At some point, someone changed a definition and I had to add an #ifdef to import addrinfo.h for VC++ 6.0. I noticed the same had to be done for Cygwin and there was a similar GCC entry. Each checkin was done by a separate person fixing a crash in their own part of the universe. The right answer would have likely been to refactor the original change that mucked up everyone else's compile. Instead we ended up with a pile of conditionals that will be hard to maintain going forward because the surrounding code has become so stiff.

Of course the problem here is that it's takes a lot of courage to make changes that affect systems you don't have access to. One of the goals of Include/pyport.h is that this localizes the mess: think of a sensibly named preprocessor symbol (PY_WE_NEED_ADDRINFO_H?) and put all the bullshit needed to work out whether to define it in one place. But you really want to get things like this right, especially if one of the affected platforms is something a little obscure. Break things on linux/x86 and you'll find out pretty quickly, but break them on Irix and it may take months and it won't be obvious which change is at fault.

Still, I want at some point to do test builds on all the test drive machines at some point before 2.4 final, so any changes will hopefully get some amount of testing on these systems at least.

Cheers, mwh

-- (Unfortunately, while you get Tom Baker saying "then we were attacked by monsters", he doesn't flash and make "neeeeooww-sploot" noises.) -- Gareth Marlow, ucam.chat, from Owen Dunn's review of the year

More information about the Python-Dev mailing list