[Omake] [Ping Jason] Some times Shell.+ threads end up running the main build loop!

[Aleksey Nogin]

Hi,

I started to get some idea of why my MetaPRL build was "blowing up" into 
hundreds of processes under certain circumstances. Basically, what was 
going on is that when the prlcomp command fails, then there is a 
possibility that the corresponding OMake thread would start executing 
its own copy of the main build loop instead of exiting! For some strange 
reason this only happens when the -S option is enabled (100% 
correlation), so passing VERBOSE=1 (without adding an explicit "-S") 
would cause the problem to disappear! Also, I was running all my tests 
with "OCAMLFLAGS=-p -j10" in the environment (and I would guess that 
these are also needed for reproducing the problem).

Note that the prlcomp (in mk/prlcomp) is currently implemented as:

#
# XXX: JYH: this seems kind of inefficient to me.
# We should fix omake so that the final line can
# just be $(argv).
#
Shell. +=
    prlcomp(argv) =
       setenv(MP_ROOT, $(ROOT))
       setenv(MPLIB, $(LIB))
       setenv(MP_GROUP, $(THEORYNAME))
       setenv(MP_DESCR, $(THEORYDESCR))
       setenv(MP_INCLUDE, $(concat :, $(OCAMLINCLUDES)))
       code = $(shell-code $(argv))
       if $(eq $(code), 0)
          return 1
       else
          exit $(code)

and the problem, I am guessing, is that something is catching the 
ExitException in the subthread and transferring the control back to the 
main loop!

So, I guess we should probably:

- Add some code that would make sure that the control in the "Shell." 
subthreads can never escape (no matter how wrong the corresponding osh 
function is)

- Fix the above "XXX".

P.S. I was doing all my tests in latest 0.9.8.

-- 
Aleksey Nogin

Home Page: http://nogin.org/
E-Mail: nogin at cs.caltech.edu (office), aleksey at nogin.org (personal)
Office: Moore 04, tel: (626) 395-2200