On Nov 2, 2012, at 1:27 AM, Fredrik Öhrström wrote: > 2012-10-26 17:02, Kelly O'Hair skrev: >> I tend to disagree, having multiple words in CC is unconventional, and if we >> do that, we need to consider the consequences of straying from the typical >> make convention. But I'd rather not get into what is theoretically wrong or >> right, it just needs to work, so we do what we have to. The fixpath trick >> concerns me on windows, all the extra exec's and all, but I've kept quiet >> about it because I'm not that up to speed on the specific issues. I think we >> need to watch fixpath pretty closely, something tells me we aren't done with >> this issue, just a gut feeling. > > Since the CC specified in spec.gmk is not shared with any external makefile, > and all implicit or default rules are turned off (using -R) > it does not really matter that it does not contain a single executable.
What about the hotspot makefiles? That's where the issue came up. We passed our convention down into hotspot makefiles. > > It would be trivial to create a CC_PREFIX variable and change all occurrences > of CC to $(CC_PREFIX) $(CC), and the same for CXX, LD etc etc > But I am not sure this would help the cygwin/msys file system confusion. It's the developer that wanders into our Makefiles and tries to make changes based on the typical GNU makefile conventions that I worry about. Granted they will eventually figure it out, I just had hoped we could avoid breaking old conventions if we don't have to. Not a big deal, we have fatter fish to fry right now. ;^) > > fixpath does not incur the cygwin fork cost, because it is a windows only > program and uses CreateProcess, not a fork simulation. And the cl.exe is a windows program too. So we have doubled the number of CreateProcess's. Windows does not manage the process memory as well. Sorry, I just get paranoid around Windows build stability, so for now, ignore the paranoid guy in the corner. ;^) -kto > > //Fredrik > >
