[BUG] Possible hang of make with tricky use of BUILT_SOURCES (was: Re: [PATCH 0/5] build: refactoring and preparations for Automake-NG)

[Stefano Lattarini]

[Adding the Automake-NG list in CC]

On 08/21/2012 08:37 AM, Paolo Bonzini wrote:
> Il 20/08/2012 21:56, Stefano Lattarini ha scritto:
>> After these changes, the Smalltalk build system still works with
>> mainline Automake (and only with it), but will be much easier to
>> modify to convert it to Automake-NG.
> 
> Can you send a "patch 6" with the remaining changes?
> 
> I'm a bit scared of this:
> 
> * libgst/Makefile.am ($(srcdir)/match.stamp): No need to specify
> the '$(srcdir)/' explicitly, since the Smalltalk build system is
> assuming GNU make anyway (it was once necessary to avoid a bug
> with FreeBSD VPATH handling); so rename this target simply ...
> (match.stamp): ... like this. Also, depend on 'genbc$(EXEEXT)',
> rather than calling '$(MAKE)' recursively in the recipe to build
> it.  Not only is this more efficient, it also avoid possible
> hanging of "make all" in the 'libgist' subdir when Automake-NG
> is used to bootstrap the Smalltalk build system.
> 
> Why should it hang "make all"?
>
Because of a limitation in how BUILT_SOURCES are handled in
Automake-NG.  This limitation is the price to pay to allow us
avoid an extra recursive make invocation per each Makefile
using BUILT_SOURCES -- and that's definitely a price worth
paying; however, I agree it shouldn't silently cause hanging
or fork bombs; we should detect and diagnose the situation.
I'll try to cook something up in the next days.

> Isn't it a bug in Automake-NG?
>
It is just a limitation.  But it is a bug that the breakage
is silent instead of being diagnosed.

Thanks,
  Stefano