Re: Fix doc index name on Windows

[Jonathan Larmour <jifl@eCosCentric.com>]

Eli Zaretskii wrote:
>>From: Pedro Alves <pedro@codesourcery.com>
>>Date: Fri, 26 Nov 2010 12:03:05 +0000
>>Cc: Jonathan Larmour <jifl@ecoscentric.com>
>>
>>On Friday 26 November 2010 11:57:47, Eli Zaretskii wrote:
>>
>>>Then it's a bug in the cross-build version of makeinfo.  There's code
>>>in makeinfo/node.c:cm_node to handle the case when a file name
>>>produced from a node name clashes with a name of an existing file
>>>(produced from some other node name), due to limitations of the
>>>underlying filesystem.  What makeinfo does in that case is put all the
>>>nodes whose names map to the same file name on that single file.  I
>>>see this behavior in action in the Windows port of makeinfo 4.8, and
>>>the code which does this was written long before Texinfo 4.7 was
>>>released, so you must have it as well.
>>
>>If such code is only triggerable on some hosts only, then IMO it
>>is broken, because the resulting files will not be movable between
>>hosts (e.g., generate on Unix, unpack on Windows/NTFS/FAT/Samba, whatnot).
> 
> 
> IMO, "broken" is an exaggeration.  How many tools did you see that
> care about having their files produced on Unix be compatible with
> NTFS?  How many maintainers of GNU packages do you know who would even
> consider a possibility of inserting NTFS-related limitations into
> their codebase?

Evidently quite a few because my patch just brings GDB into line with what 
GAS, binutils, BFD, CPP, and LD's docs explicitly do. In this respect GDB 
is the odd one out.

> Usually, such problems are at best fixed for the hosts that use the
> affected filesystems.  And the cross-build environments should take
> care of these issues, because they _do_ (or should) care.

People don't use a special cross version of makeinfo. GDB doesn't try and 
invoke one, for example. Which is quite right. I don't think people would 
be keen if I submitted a patch requiring cross builders to have provided 
e.g. arm-eabi-makeinfo.

>>Is there a way to force that behaviour with a makeinfo command line
>>> switch or something of the sort?
> 
> 
> Not that I know of.  No one has ever asked for that, AFAIK.  But it
> should be trivial to add such a switch, now that I pointed to the code
> which does that.

I don't think it's good to make GDB dependent on a version of makeinfo 
which does not yet exist.

Jifl
-- 
eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
Registered in England and Wales: Reg No 4422071.
------["Si fractum non sit, noli id reficere"]------       Opinions==mine