[Info-vax] C99 updates to CRTL

[Craig A. Berry]

On 7/30/19 2:41 PM, John Reagan wrote:
> On Tuesday, July 30, 2019 at 2:20:41 PM UTC-4, Simon Clubley wrote:
>> On 2019-07-29, John E. Malmberg <wb8tyw at qsl.net_work> wrote:
>>>
>>> The only two issues with include files that are impacting me that I
>>> currently remember are:
>>>
>>> 1. GNU software expects "../foobar.h" to work.
>>> 2. If Decnet IV is installed include <net/if.h> does not work.
>>>
>>
>> Perhaps John or someone else from VSI can comment on whether the
>> first one is supported with clang on x86-64 ? I am actually surprised
>> support for it isn't available in the compiler at the moment, because
>> that's a common relative path description in Unix software I come
>> across on Linux.
>>
> 

> Well, since I don't have anything running on x86 yet, I can't tell
> you for sure. :)
> 
> The interpretation of relative pathnames in the filespec isn't really
> a compiler issue, but a CRTL/RMS issue.

#1 works fine now if you specify your include directories in the same
syntax as what's in the #include directives.  For example
cc/include="../../" or similar.  If you specify cc/include=[--] then of
course [--]../foobar.h is a mongrel path that can't be parsed by anything.

In principle, the compiler could try it, and if it gets an invalid
filespec failure, try to convert the include directory to Unix syntax
before prepending it to what's in the #include directive and trying
again.  Dunno whether that's a good idea -- the order of include
processing is already pretty complicated given the /SYSTEM_INCLUDE and
/INCLUDE qualifiers on the command line plus the various logical names
that control include processing.