Re: Building Matreshka on Windows

[Simon Wright <simon@pushface.org>]

"Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:

> On 03/11/2016 16:32, Olivier Henley wrote:
>> On Thursday, November 3, 2016 at 4:18:07 AM UTC-4, Dmitry A. Kazakov wrote:
>>> BTW, gprbuild can C. Granted, gprbuild is a big step back from gnatmake,
>>> but still miles ahead of make.
>>
>> Why do say gprbuild is a big step back from gnatmake? I don't have
>> enough knowledge of both to evaluate your saying.
>
> Because in my opinion it is built upon an inferior technology. It uses
> XML, it is not self-contained (has a mess of semi-hidden supplementary
> files spread all around the system). It is very easy and common to
> have gprbuild not working where gnatmake never had any problem. The
> biggest danger is turning it into yet another "configure" / CMake.

Unless you go out of your way, the configuration files are all under
$prefix/share/gprconfig/

When things work, all is fine. When they don't, it can be tricky; the
way the configuration files specify commands to run and grepping the
output for particular patterns is challenging:

On the Mac, GCC 5 and 6 generate shared libraries with names like
libgnat-6.dylib; GCC 4, names like libgnat-4.8.dylib. gnatls reports
itself as e.g. GNATLS 6.1.0. The original share/gprconfig/compilers.xml
file looked for the first 2 fields of the version; libgnat-6.1.dylib
wasn't found. The fix was (cut back not to bore you)

     <version>
       <external>${PREFIX}gnatls -v --version</external>
-      <grep regexp="^GNATLS.+?(\d+(\.\d+)?)" group="1"></grep>
+      <grep regexp="^GNATLS.+?(\d+)" group="1"></grep>
     </version>

I'm very pleased that, since GPRBUILD 2016, I can use e.g.

   for Target use "arm-eabi";
   for Runtime ("ada") use "ravenscar-sfp-stm32f4";

in the GPR instead of having to specify on the command line.