Bruno Haible <bruno <at> clisp.org> writes: > Now that you say it, it's obvious what is going on: the code has first > decided that it doesn't need the 'a'/'A' substitute (since Cygwin already > has it) and then decided that it needs to emulate 'A'/'A' (because Cygwin > either crashes in low-memory situations or has a crippled handling of the > precision).
For clarification: cygwin 1.5.24 can (probably) crash on low memory situations (I have not tried to provoke such failures, and ulimit doesn't work on cygwin, so it would involve a gdb session to fake out malloc/errno). I say this, because I have fixed some memory handling bugs that were in newlib at the time cygwin 1.5.24 was released. Furthermore, cygwin 1.5.24 does not support 'a'/'A'. My bug report was with CVS cygwin (which will become 1.7.0); there, I have added 'a'/'A' support, and hopefully plugged all the bad memory cases. And I can certainly vouch for my handling of arbitrary precision with the 'a'/'A' code. But since ulimit does not work on cygwin, the configure check pessimistically guesses no (and that is the correct guess as long as cygwin 1.7.0 is unreleased). > > Thanks for the single-stepping. I'm committing this, which will hopefully > fix it (untested). I'll give it a spin. Assume success unless I send a followup mail. -- Eric Blake
