Indeed, it looks like you're not hitting the space-in-path problem. Great!
I hacked my training install to pass the dir argument along the lines you described in an earlier email, but by a different method. The problem I had that required this was that some of the embedded syscalls inside some of the perl scripts weren't keeping track of the current working directory very well. Some of them weren't passing environment vars, and I couldn't kick the problem without the hack to pass the absolute path. It's probably the same problem you're having. I was going to submit my hacks as a patch, but I think I'm too many versions behind at this point for them to still be viable. I tried to take the relative path route, but plenty of the scripts failed when I did so. I think this is the same issue you're facing. I've pretty much given up on training inside of cygwin, except for toy problems. I've found my cygwin installation to only be of use during final end-user-style decoding. MERT required too many cpu hours for me to get training to be truly viable in an all-windows world. Without the parallelism of SGE, it took far too long for me to train. I now do all my training on a SGE-linux environment. I suppose that if there were a good SGE-compat. grid environment for cygwin and I had a bunch of machines hooked up to it, I'd reconsider. This doesn't fix your situation at all, but perhaps the commiseration can lend you comfort. I'll submit a patch to you and Hieu containing my diffs from the version I have checked out. Maybe it'll help. -John On Feb 18, 2008 2:27 PM, J C Read <[EMAIL PROTECTED]> wrote: > I don't believe this to be the problem in my case. Under the Windows file > system > the path to the scripts is > C:\cygwin\home\test\moses\moses\bin\moses-scripts\scripts-20080215-1500 > > As you can see there are no spaces. In any case bash when called from the > command line behaves nicely when the scripts are called with absolute paths. > Why should it behave differently just because bash is being invoked from a > perl > script? My understanding was that programs called under cygwin can only see > the > unix-type directory hierarchy under cygwin and that as far as any program > called is concerned the root directory of cygwin is the root directory of the > file system. > > One thing I have noticed from the scripts is that commands like 'rm', 'mkdir' > etc. are called and executed without error. These executables can only be > found > with the $PATH variable. > > I wonder if using relative paths to the scripts or adding the path to the > scripts to the $PATH variable and calling the scripts by name would work? > > I think it's worthwhile finding a workable fix for this that will make moses > run > under Cygwin out-of-the-box for all users. > > > Quoting John Henderson <[EMAIL PROTECTED]>: > > > J C - > > > > I've had luck with this process, but it involved working in a > > directory without spaces in the path. I make a /sandbox root-level > > dir in cygwin and work there. It's very hard to get bash to play nice > > with paths with spaces in them. The default home directories in XP > > usually have the "Documents and Settings" directory in them. It > > causes trouble. > > > > Hope this helps. > > > > -John > > > > > > On Feb 18, 2008 9:02 AM, J C Read <[EMAIL PROTECTED]> wrote: > > > Using manually created output from step 3 of the training process I can > > confirm > > > that step 4 of the training script runs under Cygwin without any problems > > and > > > creates the two lexical files model/lex.0-0.f2n and model/lex.0-0.n2f > > > > > > Training step 5 runs partially but fails and, again, the finger of blame > > falls > > > on the safesystem calls in the extract_phrase function. Errors are similar > > to > > > those in step 3. The cygwin sh implementation doesn't seem to be able to > > find > > > these files with absolute paths when called through the perl system() > > > function. > > > > > > Don't really have any experience of working with perl under cygwin, so I'm > > not > > > sure why this is happening or how to fix it. But all experimentation seems > > to > > > point towards a default install of Moses working out-of-the-box under > > Cygwin if > > > we fix this minor issue. > > > _______________________________________________ > > > Moses-support mailing list > > > [email protected] > > > http://mailman.mit.edu/mailman/listinfo/moses-support > > > > > > > > _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
