Re: 3.14.8 / cygwin dependency

[Ralph Lange <Ralph.Lange@bessy.de>]

Hi Jeff,

It's even worse:

Obviously I'm running into filesystem cache race conditions (??)

With the $(wildcard) done by make, I get some cases where (that's my 
interpretation) make runs the wildcard before the file has been fully 
created. I'm printing the command line for the install script and 
there's no filename, just the target directory. The install script 
complains and stops the make run. When I look into the directory, the 
file is there. Running another make perfectly installs the file.

When I remove the ($wildcard) function in RULES_BUILD and directly use 
the (traditional) $<, everything works fine.
So right now I'm successfully experimenting with "$(sort $< $(wildcard 
$<*))" to force mentioning the real target once for the install script 
line. Weird!

You probably didn't see it because you're using cygwin, where the shell 
(or the cygwin dll used by cygwin make) does the wildcard expansion.

Ralph

ps. We probably should have a set of VMware virtual machines to download 
from the APS server with different Windowses and compilers readily 
installed - creating a set of comparable Windows reference compile 
hosts. The VMware player (that you need to run these virtual machines on 
Windows or Linux) is free to use. Would be a very easy-to-use and 
flexible way to run test compilations on a variety of systems without 
having multiple hardware. Creates a license problem for Windows guest 
systems, I guess. Not much of a problem for most Linux guests, though.


Jeff Hill wrote:

> Thanks Rlaph. 
>
> Actually surprised that I havent seen this here.
>
> Jeff
>
>  
>
> > -----Original Message-----
> > From: Ralph Lange [mailto:Ralph.Lange@bessy.de] 
> > Sent: Monday, January 23, 2006 7:47 AM
> > To: Janet Anderson
> > Cc: EPICS Core Talk
> > Subject: Re: 3.14.8 / cygwin dependency
> >
> >
> > Hi Janet,
> >
> > I found the change that broke the non-cygwin build:
> >
> > In RULES_BUILD 1.88.2.14 the line to install binaries was changed from
> >   @$(INSTALL_PRODUCT) -d -m 555 $< $(INSTALL_BIN)
> > to
> >   ifdef WIN32
> >       @$(INSTALL_PRODUCT) -d -m 555 $<* $(INSTALL_BIN)
> >   else
> >       @$(INSTALL_PRODUCT) -d -m 555 $< $(INSTALL_BIN)
> >   endif
> >
> > Immediately afterwards the wildcard was taken back for 
> > BORLAND compilations. The problem is that the standard 
> > Windows shell - or whatever make uses 
> > to call programs on non-cygnus systems - does not allow wildcards in 
> > this place. Changing the "$<*" to "$(wildcard $<*)" lets make do the 
> > wildcard job, which should be the better option, anyway.
> >
> > What exactly were you trying to "catch" with that wildcard? For my 
> > Microsoft non-cygnus system I actually can't think of any other files 
> > that would be matched by it.
> >
> > I didn't commit any changes to CVS as I wasn't sure about the 
> > usefulness 
> > of that wildcard.
> > So - if we need the wildcard, please change it as shown above, which 
> > will work everywhere. If we don't need the wildcard, going 
> > back to the 
> > traditional line is fine with me.
> >
> > Cheers,
> > Ralph
> >