I wonder if a better internal data structure for paths is not actually as full paths but as path parts in a list - I realize that what I'm saying is not a trivial rewrite but I think in terms of a lot of stuff having to fancily handle handoffs of posix-to-posix, posix-to-windows, windows-to-posix, and windows-to-windows whether it might ultimately simplify a couple pieces of the code that are pretty nasty. I have not fully traced this down yet, and I am not going to do so anytime soon.
What seems to be even worse, is that when I test saved a case similar to above with the "Cluster computers will be running Windows" option, unless the decoding happens differently depending on the OS (which is possible since I haven't tried any of this on Windows yet), I get some\\path\\to\\g\\etc - aka we've now lost the separator entirely just before the Metadata tag. If we lived in a purely posix world, we'd just check for \\g and then, if it isn't there, \g (either in apply_metadata and/or alter_url_post_create_batch), but we don't. This means that when, in the the modules themselves trying to figure out where to save, get_absolute_path passes something to apply_metadata, the stripping of double backslashes strips the piece in question to just g without its backslash at all, which is why we aren't seeing conversion. In a purely "posix-to-posix" situation, I'm reasonably certain that the source of the issue is that in alter_path_to_create_batch, the url2pathname converts some/path/to/\g/etc to some/path/to/\\g/etc this is not getting fixed in alter_url_post_create_batch, though one could argue maybe it's a good place for dropping such a fix. So I've traced this down somewhat, but it's gonna be messy to fix, at least in terms of "small spot fix" - basically, there's at least a couple of things going on. Stretch goal 1 is "passing in -f -l but all the files in there have identical metadata values for the key in question" works, stretch goal 2 is "everything works".ĭesktop (please complete the following information): Then, try to execute it headlessly with cellprofiler -c -r -p path/to/batchfile.h5 -o path/to/output/folder -f 1 -l 1 or cellprofiler -c -r -p path/to/batchfile.h5 -o path/to/output/folder -g Metadata_Position=POS002 batchfile_metadata_Īt minimum, passing in a -g that matches the piece of metadata used in the subfolders should work. Make a batch file on your machine with the "welcome screen pipeline" images and the attached pipeline. I initially thought it might be caused by passing -f -l flags to the batch file when executing the individual jobs, but I've now confirmed it seems to be happing with -g single group flags as well. It does NOT seem to be working though if you make a batch file and then try to headlessly execute - you essentially don't get the key-value pair substitution so instead of subfolders A01, A02 you literally get subfolder \g. It works in GUI or headlessly - if I have a saved cpproj set up this way then execute it with cellprofiler -c -r -p /path/to/file.cpproj -o /path/to/output, I get the subfolders that I expect. Briefly, various file saving modules in CellProfiler allow you to set that you want to save the output to "Default Subfolder" then let you put in the name of a piece of extracted metadata in the subfolder path as a "key", which then is replaced with the "val" upon output, which is useful.
0 kommentar(er)
