Re: [Patch 1/3] Small DOS issues

[Tim Van Holder]

On Sun, 2002-03-03 at 23:30, Akim Demaille wrote:
> >>>>> "Tim" == Tim Van Holder <address@hidden> writes:
> 
> >> | + if LFN=n test -f "$cache_file"; then
> >> 
> >> Nia?  This lacks a documentation in `test' in autoconf.texi, to say
> >> the least.  What is this horror?  How come Bash is not smart enough
> >> to handle this by itself?
> 
> Tim> It's a Win2K issue.  It does not properly detect devices (such as
> Tim> the NUL device) if the LFN interrupts are used.  The LFN=n
> Tim> disables the use of the LFN API in DJGPP, so this works again.
> Tim> This'll just stay local to the DJGPP port - no biggie.
> 
> :)
> 
> Tim, I'm fine with it, provided you encapsulate this horror into
> something of higher level.  Lemme ask a question: why don't you
> suggest that we LFN=n;export LFN at the top of configure?  What would
> it imply?

It would imply that anything following that is limited to 8+3, even
on OSes that support longer names (Win95-WinXP).  So basically, DJGPP
will behave (sort of) like plain old DOS.  Not the cleanest solution at
all.
I'd rather leave it out entirely, as it is not a breaking issue (it
would merely mean that configure scripts print a warning at the start
when run using a DJGPP bash on Win2K and WinXP and no cache file is
requested).