Hi all, I'm running into a bug but I'm not sure if it's properly considered a bug in pkg-config, or in a widely used aclocal.m4 test, or (I believe) both. I wanted to run this past the developers and see what you think. 1) The bug in pkg-config is that it expects a command line of the form: $ pkgconfig --atleast-pkgconfig-version=0.9.0 but silently accepts (and misparses) a command line of the form: $ pkgconfig --atleast-pkgconfig-version 0.9.0 Consider the following. All but the last invocation produce the expected result. The last invocation produces an unexpected result, and is arguably a bug. $ ./pkg-config --version 0.23 $ ./pkg-config --atleast-pkgconfig-version=0.24.0 $ echo $? 1 $ ./pkg-config --atleast-pkgconfig-version 0.24.0 $ echo $? 1 $ ./pkg-config --define-variable=0=0 --atleast-pkgconfig-version=0.24.0 $ echo $? 1 $ ./pkg-config --define-variable=0=0 --atleast-pkgconfig-version 0.24.0 $ echo $? 0 Digging further indicates that the first three invocations call compare_versions("0.23", "0.24.0") returning -1, whereas the last invocation calls compare_versions("0.23", "0=0") returning 1. 2) The bug in aclocal.m4 tests is simply that it invokes --atleast-pkgconfig-version with the wrong syntax. This is widespread, almost everywhere pkg-config is used by configure, in the PKG_PROG_PKG_CONFIG macro. I ran into this problem doing something like this: PKG_CONFIG="/usr/bin/pkg-config --define-variable=prefix=/foo/bar" ./configure ... which does not function as expected and I believe is a perfectly reasonable thing to expect to do with pkg-config and configure! What do you think? Thanks, Martin