svn commit: r277803 - projects/clang360-import/lib/clang/include

[Dimitry Andric]

On 28 Jan 2015, at 20:24, Tijl Coosemans <tijl at freebsd.org> wrote:
> 
> On Wed, 28 Jan 2015 13:39:46 -0500 Alexander Kabaev <kabaev at gmail.com> wrote:
>> On Wed, 28 Jan 2015 16:37:44 +0100
>> Dimitry Andric <dim at FreeBSD.org> wrote:
>>> On 28 Jan 2015, at 15:14, Alexander Kabaev <kabaev at gmail.com> wrote:
>>>> On Wed, 28 Jan 2015 08:42:48 +0100
>>>> Dimitry Andric <dim at FreeBSD.org> wrote:
>>> ...
>>>>> I'm not sure what the problem is with storing a compiler's
>>>>> internal-only headers in the location where upstream expects them
>>>>> to be?  Note that gcc does something similar; for example with the
>>>>> gcc49 port, it stores all its internal headers under:
>>>>> 
>>>>> /usr/local/lib/gcc49/gcc/i386-portbld-freebsd11.0/4.9.3/include
>>>>> 
>>>>> While I do agree that this is not a pretty-looking path, upstream
>>>>> has chosen it to be like this, and there are most likely good
>>>>> reasons for it.  As for clang, I just want to get rid of as many
>>>>> "FreeBSD is a special snowflake" patches as I can.
>>>> 
>>>> Nobody _expects_ them to be there, for they are internal and not
>>>> directly addressable by anyone. We can tweak locations as we see fit
>>>> with no user visible consequences. What you destroy by this is the
>>>> nice property we had to date - all of the base headers could be
>>>> searched by simple grep on /usr/include and no obscure directories
>>>> need to be remembered.
>>> 
>>> First you say "they are internal and not directly addressable", and in
>>> the next sentence you say that it is a "nice property that you could
>>> search by grepping /usr/include".  If the files are internal and not
>>> directly addressable, why would you ever want to search them?
>> 
>> Because people do software development on FreeBSD from time to time
>> they tend to want to know what's in there. If one has to resolve to
>> start hunting for answers in header files, all they want is for you to
>> start hiding them in random places. grep -r <blah> /usr/include is easy.
>> 
>>> In that respect, it is even better to relocate them to a different
>>> location than /usr/include, since they're not *FreeBSD* headers,
>>> they're clang internal headers.
>>> 
>>> In other words, I consider this a win, not any form of loss. :)
>> 
>> They are FreeBSD headers as long as you ship clang as part of FreeBSD
>> sources. Your change does not improve usability, since headers are
>> internal and having them under /usr/include does not hurt anyone nor is
>> that a change that is hard to maintain. By moving them you actively
>> hurt source transparency by forcing someone to potentially hunt for
>> answers in many disjoint directories and as such this is a net loss.
>> No smilies.
> 
> FreeBSD doing things slightly differently than everybody else is often
> also a cause of trouble for developers.  In this case too.  The fact
> that FreeBSD didn't have /usr/lib/clang/$VERSION/include changed the
> output of "cc -print-file-name=include" which is why there's an ugly
> /usr/lib/include symlink.  This link can be removed now.
> 
> That said, this commit breaks compiling with "cc -ffreestanding
> -nostdinc -I`cc -print-file-name=include`" because clang versions of
> std*.h headers aren't installed on FreeBSD which is yet another
> thing that FreeBSD does differently.  Maybe now is a good time to fix
> that as well?

We never installed any internal clang headers that were named the same
as our own headers, because they used to cause conflicts.  I will have
a look if those conflicts have now been resolved, then we could install
them for use in freestanding environments.

-Dimitry

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 194 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://lists.freebsd.org/pipermail/svn-src-projects/attachments/20150128/0901c1ab/attachment.sig>