cygwin build_alias issue (and a possible workaround...)

[Chris Johns]

Bogdan Vacaliuc wrote:
> 
> Thanks for that data point.  I guess there is still a deficiency.
> Please try this (since it is easy for you, I think):

I have access to the machine at the moment.

> Add a function using the option=value items from your test-fail script into the test-configure script.

I have and I attach the script for you plus the fail.log (appended 
together). Note, all the failing tests on WinME pass on WindowsXP with 
MSYS. More below ...

   [ Note, I needed to remove some '-' characters from the functions
     for the MSYS sh to run the script. ]

> I would expect the failure to occur on the autoconf generated configure more-or-less
> with the same frequency as when you call your test script.

It does. It fails on the first pass on WinME.

> 
> [NOTE: I had to remove all (3?!) references to --srcdir=, but other than that it's the
> same as your command line in test-fail]
> 
> I have observed sensitivity to kind and quantity of arguments as well; perhaps the
> build system stirs in additional entropy? 

I agree. I have added a number of new tests I hope will narrow the 
failure down:

ccjfail
   This is the command line in the nested configure that fails for
   me on WinME and MSYS.

ccjminfail
   I removed the options starting at the first until the fail did not
   occur. Less than 4 '--' type options and the failure stopped. More
   did not seem to change the result as ccjfail shows.

ccjmin2fail
   The failure with the build_alias variable is created on MSYS on
   WinME.

ccjminpass
   Removing a '--' type option stopped the failure.

ccjmin2pass
   Same as ccjminpass except the build_alias parameter this time.

The configure script's option handling loop is a regexp 'expr' command 
to filter the option's arguments then a shell variable assignment for 
the '--cache=file=' option, then the failed '--enable-*' option. I 
placed the same commands in the 'test-configure' script before calling 
configure and there was no failure.

I also found hacking the configure script where the '--enable-*' option 
is handled stopped the failure. I added the same call to 'expr', without 
an exit code check, that tests for a valid shell script variable. 
Basically, this specific regexp 'expr' call seems to fail with expr 
returning an error code.

> It
> would be better if we could come up with a list that covers *all* machines
> and environments.  Clearly mine is less than that right now.
> 

MSYS on WindowsXP _seems_ fine, while WinME is not. I have no idea why.

-- 
  Chris Johns
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: fail.log
URL: <http://lists.rtems.org/pipermail/users/attachments/20040914/fc29112f/attachment-0002.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: test-configure
URL: <http://lists.rtems.org/pipermail/users/attachments/20040914/fc29112f/attachment-0003.ksh>