Re: Using "make -d l" makes buildworld to fail

On Mon, 5 Sep 2005, Ruslan Ermilov wrote:

RE>On Mon, Sep 05, 2005 at 05:05:47PM +0200, Harti Brandt wrote:
RE>> On Thu, 4 Aug 2005, Ruslan Ermilov wrote:
RE>> 
RE>> RE>On Thu, Jul 28, 2005 at 08:14:15PM +0200, Jeremie Le Hen wrote:
RE>> RE>> Hi,
RE>> RE>> 
RE>> RE>> I have to run multiple buildworlds with and without debugging ("-d l")
RE>> RE>> these last days.  I was nearly turning mad while I realized that
RE>> RE>> buildworld would fail in "rescue" when using "-d l" make option.
RE>> RE>> 
RE>> RE>> Both typescripts with and without the debugging option are available :
RE>> RE>> 	http://jeremie.le-hen.org/~tataz/make_-d_l_buildworld.gz (ko)
RE>> RE>> 	http://jeremie.le-hen.org/~tataz/make_buildworld.gz (ok)
RE>> RE>> 
RE>> RE>> This is not the stock source tree, but I checked on a non-modified
RE>> RE>> tree and the behaviour is the same.
RE>> RE>> 
RE>> RE>crunchgen(1) is broken: it generates a makefile to find a
RE>> RE>list of object files for a given program, then calls it,
RE>> RE>parses the output, and expects it to start with the line
RE>> RE>"OBJS=".  If something is passed in the MAKEFLAGS environment
RE>> RE>variable, like -Dl, things are really screwed.
RE>> 
RE>> Couldn't we make crunchgen just to ignore lines that don't start with 
RE>> OBJS=? Or we could finally make make(1) to spit debug output on stderr.
RE>> 
RE>The latter would be great, and would fix the problem in question.

I had a look at this but it appears make(1) already prints it's debugging
information on stderr. The only exception is -dl which prints commands 
even when they are prefixed with _at_. I would actually expect make(1) to 
behave this way. So I suppose we should find a solution for this problem 
in crunchgen.

harti