Re: [libvirt] [PATCH 01/29] Add some autoconf helper macros for checking for libraries

On 09/20/2012 09:01 AM, Daniel P. Berrange wrote: On it's own, this patch merely introduces files that are unused, but it doesn't make sense to push before the release unless we also take one of the other commits in the series that uses it. Missing a comma. Also, autoconf ignores whitespace after comma, so I'd write this as: LIBVIRT_CHECK_LIB([SELINUX], [selinux], [getfilecon], [selinux/selinux.h]) More precisely, in '-lselinux', since this construct works even for cygwin where such a library [if it existed] would be named cygselinux.dll. s/existance/existence/ Missing a copyright header. s/existance/existence/ 5 arguments listed here, but only 4 listed in the commit message. Will these two values always be the same case-insensitive string? If so, you only need one of the two of them, and use m4 to do the case-conversion as part of this macro, using functions such as m4_toupper. I'd allow this to be a default, where the user can write: [NAME_VAR], [], instead of: [NAME_VAR], [NAME_VAR], in the common case (of course, you demonstrated sanlock_client as a counter-example that warrants the separate arg). To do this, you replace use of $3 with m4_default([$3], [$2]). s/eg/e.g./ Long lines. Encourage wrapping at a sensible point. Missing quoting. You need [] around the second argument of AC_ARG_WITH. AC_HELP_STRING is obsolete. Autoconf recommends using AS_HELP_STRING. Line wrapping is hard to follow; I'd put a line break after the HELP_STRING() and before the [] third argument of AC_ARG_WITH. Technically, $with_var is user-supplied, and can therefore start with text that can confuse (non-compliant) versions of 'test'; it's better in this case to write test "x$with_var" != "xno" (here, and anywhere else a shell variable with user-supplied contents is probed). Also, here, you are directly probing with_var... ...but here you are using the AS_VAR_* wrappers. If you plan on using the AS_VAR wrappers, then all uses of the variable, including probing its value, have to go through AS_VAR (that is, $with_var might not be valid shell code compared to AS_VAR_GET([with_var]), depending on what the user gave as input to this macro). But it looks like you intend for this macro to always be used with literal strings rather than shell indirections, at which point AS_VAR_PUSHDEF/AS_VAR_POPDEF may be overkill, and you could get away with simpler m4_pushdef/m4_popdef. Technically, the "" aren't necessary here, but they don't hurt either. s/existance/existence/ A lot of autoconf macros have the style of providing two arguments, one expanded if the check was successful, and the other if it failed. I'm wondering if it would be better to have: LIBVIRT_CHECK_LIB([WITH_VAR], [NAME_VAR], [LIBNAME], [FUNCNAME], [HDRNAME], [if-found], [if-not-found]) where this macro would not be needed, but you would instead write it as: LIBVIRT_CHECK_LIB([WITH_VAR], [NAME_VAR], [LIBNAME], [FUNCNAME], [HDRNAME], [], [LIBVIRT_CHECK_LIB([WITH_VAR2], [NAME_VAR], [LIBNAME2], [FUNCNAME2], [HDRNAME])]) > +dnl > +dnl WITH_VAR: Suffix for the WITH_XXX variable in config.h & conditional in make, > +dnl and prefix for the XXX_CFLAGS and XXX_LIBS make variables > +dnl WITH_VAR2: Suffix for the WITH_XXX variable in config.h & conditional in make > +dnl if the fallback library was required > +dnl NAME_VAR: Suffix for the --with-XXX configure arg and $with_XXX configure variable > +dnl LIBNAME2: Suffix for the library name libYYY.so fallback choice > +dnl FUNCNAME: Name of function to check for in libXXX.so > +dnl FUNCNAME2: Name of function to check for in libYYY.so > +dnl HDRNAME: Name of header file to check for Why don't we need a HDRNAME2? And should we make it easy to default contents where any *2 variable left blank has the same value as the non-2 variable? > + AS_VAR_PUSHDEF([name_var],[$3]) > + AS_VAR_PUSHDEF([arg_var],[with-$3]) > + AS_VAR_PUSHDEF([with_var],[with_$3]) > + AS_VAR_PUSHDEF([libname_var],[$4]) > + AS_VAR_PUSHDEF([libname2_var],[$5]) > + AS_VAR_PUSHDEF([funcname_var],[$6]) > + AS_VAR_PUSHDEF([funcname2_var],[$7]) > + dnl AS_VAR_PUSHDEF([hdrname_var],[$8]) > + > + AC_ARG_WITH([name_var], > + AC_HELP_STRING([--arg_var], Same comments as before. Also, here, you could write: [with lib]with_var[ support ...] instead of having to look up what $3 meant. If you do use my suggestion of an if-found/if-not-found argument, then this would be better as: if test $fail = 1; then m4_default([if-not-found-arg], [AC_MSG_ERROR([...])]) fi so that failure to find the library is fatal only if the macro caller didn't supply alternate if-not-found code. Missing matching AS_VAR_POPDEF, throughout the patch. Every AS_VAR_PUSHDEF (or m4_pushdef) needs a paired cleanup within the same macro definition. Missing a copyright and documentation. Why does this need to be a separate file? printf -v is not portable. This has to be written: STR=`printf "%8s: %-3s" "$1" "$2"` -- Eric Blake eblake(a)redhat.com +1-919-301-3266 Libvirt virtualization library http://libvirt.org