-
Notifications
You must be signed in to change notification settings - Fork 10
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Decouple config filename from results filename and remove need for config file from tools which consume results. #48
Comments
I'm not sure that is so bad to name the file the way we have. The real issue is that consuming the results is tricky. I have a script somewhere that converts the results data to The Json file is really only useful to krun, because it contains the extra information needed for resume-mode and so on. |
I'm not sure I understand the CSV stuff. What I mean by bullet point 2 of this bug is that tools should not need to have both a config file and a results file present in order to work. Look at
whereas it would be nicer if it were:
Seems silly at first, but the only remaining reason the config file is passed used is to derive a results file name (as described in original report). And it imports krun just to do this (uses a function in If we were to make this change, we would then be free to name the results files as we wish, with a datestamp if we wanted. I dunno, maybe it doesn't matter. |
FWIW, I'm not much a fan of CSV as a format. I chose JSON as it's hierarchical and is easily parsed in just about any language. |
I'm a bit baffled by this. I'll have another read through the code when I've finished the current batch of commits. |
Let me try again from a more practical standpoint. Currently you run krun on a config file, say I find this annoying as it's too easy to overwrite results, as all results files (for that config) are named the same. I've overwritten results a couple of times accidentally. We could add a datestamp to the results file name (like with the log files), but we would have to rework existing tools that take as an argument a config file, and expect to be able to derive the name of the results file from this. Basically anything that depends upon Hope that helps. |
I'm neutral on this, but an alternative is for krun to say "I won't overwrite a results file unless you specify --force" or similar. |
It sort of helps. A simple change I can add in to my current commits would be to pass the result filename in as an argument to krun:
If we make the change above the mk_graphs script will be slightly more broken, but since it already has a bug we can deal with that all in one go. I still don't quite understand why the current tools need the config file, because the config is stored in the results file. So surely they can just read the config from the dumped results? |
The tools no longer need anything from the config (in either the config file, or the config stashed in the results file). So they should not depend upon the config file at all. The config file in the results file is only there for future reference, should we wish to know what parameters were used. |
On the flipside, we may find a need for a tool to require something from the config file later... |
True. Do you want me to add an |
I'm no longer sure what is best, but some part of me thinks the result files should be date-stamped so as to match the log filename. |
In which case I won't act on this now; we can come back to it another time. |
Currently the result filename is derived from the config filename. E.g.:
Tools that consume results use a function in krun to discover the json filename.
I think it would be better if:
warmup_20151008.bz2
This is really just an annoyance and isn't all that high priority.
The text was updated successfully, but these errors were encountered: