Den 2010-08-16 21:41 skrev Peter Rosin: > Den 2010-08-16 21:36 skrev Eric Blake: >> On 08/16/2010 01:26 PM, Peter Rosin wrote: >>> Good catch, I knew they were very similar and assumed my matching was just >>> about wide enough. But since I didn't check, Murphy made sure that it wasn't >>> working for everybody, see below... >> >> Rather than doing a sed for a particular regex, I would be much more >> comfortable with a comparison of the two outputs. Less chance of >> Microsoft changing their output in such a way that would foil yet >> another regex. > > Yes, cmp would be nicer. But the current autoconf code is not at all > rigged for doing that easily. It's way beyond me anyway, and I dare not > attempt such an overhaul...
I looked a bit further, and found that there are other tests that will also choke on the logo on stderr. AC_PATH_XTRA has no chance of succeeding as it has ac_[]_AC_LANG_ABBREV[]_werror_flag=yes before testing and never even tries with the flag cleared. Not that I'd think AC_PATH_XTRA would have much success on MSVC anyway, but more instances of *_werror_flag=yes are bound to creep in. Heck, there's even a published macro for setting it, who knows what 3rd party code might do with it turned on? So, a cmp on stderr in the -g test will not solve all issues with the boilerplate logo on stderr, but my patch will. That is, if you ignore that it's a bit over on the fragile end of the scale and might need adjustment in the future. On the other hand, someone probably thought the current -g test was pretty solid at some point and it still needed adjustment... Cheers, Peter