autoconf
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Testing for unknown flags in different compilers


From: Eric Blake
Subject: Re: Testing for unknown flags in different compilers
Date: Fri, 07 Mar 2014 14:17:39 -0700
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.3.0

On 03/07/2014 02:07 PM, Bob Friesenhahn wrote:
> On Fri, 21 Feb 2014, Eric Blake wrote:
> 
>> This will still remain true.  And the developer's choice of CFLAGS
>> should always take precedence over any default flags that autoconf
>> probes for in the absence of a developer preference.
> 
> This statement is one that I definitely believe in but it makes the
> grave assumption that it is possible to undo/override options specified
> in left to right order.  This works for GCC but definitely not compilers
> at large.  How would compilers be supported which do not provide the
> ability to remove an already supplied option?

Probably how we already handle non-default CFLAGS:

If CFLAGS is unspecified, we choose defaults for you (currently '-g -O2'
for gcc and '-g' for other compilers - this proposal adds several
'-W...' for gcc and various other tested-to-work options for other
compilers).  But if you specify CFLAGS we honor your choice completely
(and don't try to inject our defaults)

Maybe there's some fine-tuning involved - we compute the default
warnings, and stick it on the left of CFLAGS even if you explicitly
override CFLAGS, but provide yet another envvar or command line option
to disable all warnings probes; although for warnings to work by default
this secondary variable has to eventually feed CFLAGS (because any other
solution would require using a new variable in makefiles, which means
its no longer a default-on improvement to the toolchain).

There's also two levels of defaults - one is whether configure files
should start probing for warnings by default (configure.ac authors
should have final say on probing for additional warnings or bypassing
the warning probes altogether, in the case where they don't like
autoconf defaults), another at the end user level (configure users
should have final say on whether to trust the configure.ac choice of
CFLAGS, or to provide their own).

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org

Attachment: signature.asc
Description: OpenPGP digital signature


reply via email to

[Prev in Thread] Current Thread [Next in Thread]