build: avoid passing empty strings to build flags

[jbergstroem]

While checking the return values from icu-i18n, we didn't validate the content (enough) before passing it to the build system. Also make cflags parsing more robust by avoiding empty strings.

The reason we're introducing filtering is to avoid empty strings which might occur for other reasons (I found deviations between the systems I've been testing on and the stuff from arch linux). When you pass input to .split() it stops doing the "convenience stuff";

>>>'-I/tmp/foo -I/bar/baz -I'.split('-I') ['', '/tmp/foo ', '/bar/baz ', ''] >>>filter(None, _) ['/tmp/foo ', '/bar/baz ']

Fixes: #1787

/R=@bnoordhuis, @alferpal

[bnoordhuis]

Long line.

[bnoordhuis]

Aside, the filter step removes empty lines but not blank lines. Would that matter?

[jbergstroem]

I haven't come across that just yet. I'll improve the filter; guess it couldn't hurt. Just want to avoid making parsing too complex.

[jbergstroem]

@bnoordhuisfilter(False, foo) would do, right (captures '' or None)?

[bnoordhuis]

I don't think filter() accepts False. What about this?

cflags=default_cflagsifpkg_cflags: cflags=filter(False, map(str.strip, pkg_cflags.split('-I'))) # Maybe easier to read:cflags=map(str.strip, pkg_cflags.split('-I')) cflags= [sforsincflagsifs]

[jbergstroem]

Adapted a similar strategy.

[rvagg]

lgtm

[jbergstroem]

@bnoordhuis looks good to you?

[bnoordhuis]

LGTM

[thefourtheye]

Minor nitpick. What if pkg_cflags is a blank string with more than one whitespace characters? cflags would become an empty list. Is that okay?

[jbergstroem]

@thefourtheye I've never been able to get pkg-config give me a similar output. Even if it did, calling strip on " " (two spaces) would return empty:

>>>" ".strip() ''

[jbergstroem]

Committed in c5a1009. I'll drive including this in all minor releases that needs a fix. Thanks for the input and reviews everyone.