[isabelle-dev] Up-to-date instructions for 'isabelle afp_build -A'

[Gerwin.Klein at data61.csiro.au]

Good points. I’ve added the slow/very_slow and x64 hints to the documentation now.

Cheers,
Gerwin

> On 15.01.2018, at 07:38, Makarius <makarius at sketis.net> wrote:
> 
> On 14/01/18 12:42, Clemens Ballarin wrote:
>> Dear Maintainers of Isabelle / the AFP,
>> 
>> Where would I find instructions on actually getting 'isabelle afp_build
>> -A' to run through?  I was hoping to find that in
>> <afp-devel>/doc/regression-test.md but that merely states the command.
> 
> My command-line usually looks like this:
> 
>  isabelle build -d '$AFP' -a -X slow -j4
> 
> Or:
> 
>  isabelle build -d '$AFP' -a -X very_slow -j4
> 
> You can also use -D '$AFP' (without -a) to address precisely the AFP
> sessions.
> 
> 
>> It appears that I would need to set ML_PLATFORM=x86_64-darwin for the
>> large sessions but then some magic seems to defeat that and builds are
>> still for x86-darwin.
> 
> Instead of editing $ISABELLE_HOME_USER/etc/settings to change
> ML_PLATFORM (which is difficult to get right), I usually edit
> $ISABELLE_HOME_USER/etc/preferences to add:
> 
>  ML_system_64 = true
> 
> Only some "slow" sessions do need x86_64, but some non-slow sessions
> (like HOL-ODE-Numerics) become a bit slow in x86 mode. Almost everything
> else works better in x86 mode, which is the default.
> 
> See also
> http://isabelle.in.tum.de/devel/build_status/AFP_slow_64bit_6_threads/index.html
> for typical maximum heap requirements.
> 
> 
> 	Makarius
> _______________________________________________
> isabelle-dev mailing list
> isabelle-dev at in.tum.de
> https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/isabelle-dev