[VMS/committed]: Set extension to archive member

[VMS/committed]: Set extension to archive member

[Tristan Gingold]

Hi,

VMS archives member names doesn't have extension (only the 'basename' is stored).
But this is not the usual convention for ar, and some tools (libtool) expect that extracted
members of an archive to have the standard object extension.

This patch implements this behaviour, thus making VMS archives more UNIX friendly.

Committed on trunk.

Tristan.

bfd/
2012-04-17  Tristan Gingold  <gingold@adacore.com>

	* vms-lib.c (_bfd_vms_lib_get_module): Append .obj extension to
	member of an object archive.<

Index: vms-lib.c
===================================================================
RCS file: /cvs/src/src/bfd/vms-lib.c,v
retrieving revision 1.21
diff -c -p -r1.21 vms-lib.c
*** vms-lib.c	17 Apr 2012 10:28:19 -0000	1.21
--- vms-lib.c	17 Apr 2012 10:47:01 -0000
*************** _bfd_vms_lib_get_module (bfd *abfd, unsi
*** 1296,1301 ****
--- 1296,1302 ----
    struct lib_tdata *tdata = bfd_libdata (abfd);
    bfd *res;
    file_ptr file_off;
+   char *name;
  
    /* Sanity check.  */
    if (modidx >= tdata->nbr_modules)
*************** _bfd_vms_lib_get_module (bfd *abfd, unsi
*** 1357,1363 ****
        res->origin = file_off + tdata->mhd_size;
      }
  
!   res->filename = tdata->modules[modidx].name;
  
    tdata->cache[modidx] = res;
  
--- 1358,1382 ----
        res->origin = file_off + tdata->mhd_size;
      }
  
!   /* Set filename.  */
!   name = tdata->modules[modidx].name;
!   switch (tdata->type)
!     {
!     case LBR__C_TYP_IOBJ:
!     case LBR__C_TYP_EOBJ:
!       /* For object archives, append .obj to mimic standard behaviour.  */
!       {
! 	size_t namelen = strlen (name);
! 	char *name1 = bfd_alloc (res, namelen + 4 + 1);
! 	memcpy (name1, name, namelen);
! 	strcpy (name1 + namelen, ".obj");
! 	name = name1;
!       }
!       break;
!     default:
!       break;
!     }
!   res->filename = name;
  
    tdata->cache[modidx] = res;
  

Re: [VMS/committed]: Set extension to archive member

[h.becker]

On 04/17/12 12:50, Tristan Gingold wrote:
> VMS archives member names doesn't have extension (only the 'basename' is stored).
> But this is not the usual convention for ar, and some tools (libtool) expect that extracted
> members of an archive to have the standard object extension.
>
> This patch implements this behaviour, thus making VMS archives more UNIX friendly.

It's not obvious to me whether the change adds the object extension to 
the member name or if it is added just for extracting a member. Adding 
it to the member name will make the library somehow "unusal" for VMS. It 
may break tools like MMS and it sure will confuse VMS users. It may not 
break the VMS linker, but I didn't check.

For what it is worth, VMS object libraries contain object modules. The 
name of the module is determined on VAX and Alpha: by the [e]mh$t_name 
in the object module header record; on I64: by the NT_VMS_MHD entry in 
the notes section. The object language, as documented in the Linker 
manual, only restricts the length of the module name: 31 characters. It 
is up to the tool which generates the object module - in VMS terms the 
language processor - to write this module header. Usually the language 
processors allow the users to specify the module name, for example C 
with "#pragma module". (The HP C compiler expects a C identifier for 
this #pragma, which is a HP C compiler limitation). All language 
processors have default values, for the case that a user doesn't specify 
a name. Some but not all language processors use the basename. VAX macro 
for example uses ".MAIN.", which is not a valid ODS2 basename. The C 
compiler uses the basename. For ODS5 disks this can result in funny 
module names for C sources like "a.b.c".

Re: [VMS/committed]: Set extension to archive member

[Tristan Gingold]

On Apr 17, 2012, at 2:32 PM, h.becker wrote:

> On 04/17/12 12:50, Tristan Gingold wrote:
>> VMS archives member names doesn't have extension (only the 'basename' is stored).
>> But this is not the usual convention for ar, and some tools (libtool) expect that extracted
>> members of an archive to have the standard object extension.
>> 
>> This patch implements this behaviour, thus making VMS archives more UNIX friendly.
> 
> It's not obvious to me whether the change adds the object extension to the member name or if it is added just for extracting a member. Adding it to the member name will make the library somehow "unusal" for VMS. It may break tools like MMS and it sure will confuse VMS users. It may not break the VMS linker, but I didn't check.

It is added just when a member is extracted.  Therefore the library looks usual to the VMS linker (or to librarian).

> For what it is worth, VMS object libraries contain object modules. The name of the module is determined on VAX and Alpha: by the [e]mh$t_name in the object module header record; on I64: by the NT_VMS_MHD entry in the notes section. The object language, as documented in the Linker manual, only restricts the length of the module name: 31 characters. It is up to the tool which generates the object module - in VMS terms the language processor - to write this module header. Usually the language processors allow the users to specify the module name, for example C with "#pragma module". (The HP C compiler expects a C identifier for this #pragma, which is a HP C compiler limitation). All language processors have default values, for the case that a user doesn't specify a name. Some but not all language processors use the basename. VAX macro for example uses ".MAIN.", which is not a valid ODS2 basename. The C compiler uses the basename. For ODS5 disks this can result in funny module names for C sources like "a.b.c".

Although many VMS #pragma are now implemented in gcc, #pragma module isn't.  We currently set the module name from the basename of the file.

Tristan.

Re: [VMS/committed]: Set extension to archive member

[Douglas B Rupp]

On 4/17/2012 3:50 AM, Tristan Gingold wrote:
> Hi,
>
> VMS archives member names doesn't have extension (only the 'basename' is stored).
> But this is not the usual convention for ar, and some tools (libtool) expect that extracted
> members of an archive to have the standard object extension.
>
> This patch implements this behaviour, thus making VMS archives more UNIX friendly.
>
> Committed on trunk.
>
> Tristan.

So if one inserts a .o, it extracts as a .obj?

Re: [VMS/committed]: Set extension to archive member

[Tristan Gingold]

On Apr 17, 2012, at 6:23 PM, Douglas B Rupp wrote:

> On 4/17/2012 3:50 AM, Tristan Gingold wrote:
>> Hi,
>> 
>> VMS archives member names doesn't have extension (only the 'basename' is stored).
>> But this is not the usual convention for ar, and some tools (libtool) expect that extracted
>> members of an archive to have the standard object extension.
>> 
>> This patch implements this behaviour, thus making VMS archives more UNIX friendly.
>> 
>> Committed on trunk.
>> 
>> Tristan.
> 
> So if one inserts a .o, it extracts as a .obj?

Yes.  That looks strange, but the extension is lost in the archive.

Re: [VMS/committed]: Set extension to archive member

[h.becker]

On 04/18/12 08:33, Tristan Gingold wrote:
>
> On Apr 17, 2012, at 6:23 PM, Douglas B Rupp wrote:
>
>> On 4/17/2012 3:50 AM, Tristan Gingold wrote:
>>> Hi,
>>>
>>> VMS archives member names doesn't have extension (only the 'basename' is stored).
>>> But this is not the usual convention for ar, and some tools (libtool) expect that extracted
>>> members of an archive to have the standard object extension.
>>>
>>> This patch implements this behaviour, thus making VMS archives more UNIX friendly.
>>>
>>> Committed on trunk.
>>>
>>> Tristan.
>>
>> So if one inserts a .o, it extracts as a .obj?
>
> Yes.  That looks strange, but the extension is lost in the archive.
>
>
Hmm, as I tried to explain, it's a feature of VMS object libraries. The 
extension may change as well as the basename may change - because of 
differences between the module name and the basename - and as you may 
end up with many library members from just one object file.

Re: [VMS/committed]: Set extension to archive member

[Tristan Gingold]

On Apr 18, 2012, at 9:09 AM, h.becker wrote:

> On 04/18/12 08:33, Tristan Gingold wrote:
>> 
>> On Apr 17, 2012, at 6:23 PM, Douglas B Rupp wrote:
>> 
>>> On 4/17/2012 3:50 AM, Tristan Gingold wrote:
>>>> Hi,
>>>> 
>>>> VMS archives member names doesn't have extension (only the 'basename' is stored).
>>>> But this is not the usual convention for ar, and some tools (libtool) expect that extracted
>>>> members of an archive to have the standard object extension.
>>>> 
>>>> This patch implements this behaviour, thus making VMS archives more UNIX friendly.
>>>> 
>>>> Committed on trunk.
>>>> 
>>>> Tristan.
>>> 
>>> So if one inserts a .o, it extracts as a .obj?
>> 
>> Yes.  That looks strange, but the extension is lost in the archive.
>> 
>> 
> Hmm, as I tried to explain, it's a feature of VMS object libraries. The extension may change as well as the basename may change - because of differences between the module name and the basename - and as you may end up with many library members from just one object file.

Yes, I think that I understand that.
Contrary to VMS librarian, GNU ar use the filename to set member name (and not the module name).  In practice, this doesn't change anything as basename = module name.  I agree this is not fully compliant with VMS librarian, but that's UNIX tools ported to VMS.

Tristan.