Re: [PATCH] Improve portability across compilers

From: Kinkie <gkinkie_at_gmail.com>
Date: Sat, 24 Aug 2013 09:27:17 +0200

>>> * in acinclude/compiler-flags.m4 I am worried that you are leaving
>>> -Qunused-arguments present in the variable labeled Werror.
>>> - IMO this needs to be a separate flag only added if a compiler flag
>>> check
>>> proves it necessary.
>>
>> It is necessary, unless we completely overhaul our build system: the
>> "unused arguments" it refers to are unused _compiler_ arguments. Such
>> as for instance include directories added to the search path but which
>> contain no headers needed by the current translation unit.
>
>
> Clang complaining about those?
>
> Two thing to keep in mind:
> * For automake 1.13 we will shortly have to be having to remove the
> INCLUDES macro in the .am and spread its contents between LDFLAGS, CFLAGS,
> CXXFLAGS, etc properly.

That's not going to help here. Automake allows setting *FLAGS per
target, not per translation unit (and even if it could, would we want
to?)

>>> - why is the -Wno-error=unused-value being added there now too?
>>
>> I'm removing it, it's apparently unneeded.
>> Although some clang versions _are_ complainign about unused return
>> values somewhere in the CBDATA_CLASS2 macro definition. I wasn't able
>> to track where exactly so far.
>
>
> Okay. If those are older than the clang built into FreeBSD 9/10 by default
> (IIRC they were) we don't really care, they also lack C++11 functionality we
> want to make use of.

Yes, they are older (version 2.8 versus FreeBSD's 3.1). I'll leave the
jobs on in matrices but not care about results.

>>> We have a separate section for the compiler-specific default flags right?
>>> if
>>> not we need to add it.
>>
>> I'm not sure about what you mean..
>
>
> I seem to remember separate sectiosn of switch case and if statements early
> and late in the configure.ac the one you are editing here setting up the
> details for --enable-strict-error-detection variables. And a later one doing
> compiler-specific and OS-specific tuning for things like -g / -stdc++11,
> march-native, etc etc. That later flags setup was still quite messy and
> spread over a few different if statements with switches inside. (it has been
> nagging at my OCD tendencies every time I see bits of it).

It should be in the compiler-flags.m4 file, so at least be identifiable.
I would not wake the sleeping dog there for now.

-- 
    /kinkie
Received on Sat Aug 24 2013 - 07:27:26 MDT

This archive was generated by hypermail 2.2.0 : Sat Aug 24 2013 - 12:00:14 MDT