Wish list: A script transforming the .csv file generated by test/performance runtests.py into the 80-column summary that runtests.py prints. Lets call it showcsv.py for now. The showcsv.py script inputs a csv file in the format matching that of runtests.py, as well as either at most two avm nicknames used to key the entries of the csv file, and produces a 80-column summary of the performance data. Providing two avm nicknames indicates the comparative presentation should be used. (In general the .csv's I create have >2 avm nicknames, so I am not terribly interested in tricks like trying to infer the two nicknames from the file.) It would be good if showcsv.py reused the same functions that runtests.py uses for processing the test data, to guard against the introduction of inconsistent analyses as we continue to refine how we analyze and present results. There is currently not enough information in the .csv file to reliably reconstruct the entire transcript in exactly the same output format (e.g. the paths to the shells that were run and the timestamps of the runs are not included in the .csv file), but exact replication is not the point here.

The overall structure will be clearly after I post the next two patches. The only goal of this patch is to move all the class and other data definitions into its own file that can be shared by different points of entry. Q for Chris: what is "the real" Pythonic way to accomplish what I'm trying to do in this patch? Is this an abuse of import? (Is "." always on the import path?)

The purpose of this patch will be clearer after seeing the third patch that I'm about to attach. (Basically I wanted to reuse only a small snippet of the already too-large header printing code, so I refactored the large method into a collection of smaller ones.)

My main goal was for this to be short and sweet without having to refactor the runtests.py stuff *too* much. Feel free to suggest drastic changes, though I would like to stay under the 100 LOC mark. Example invocation: % python showcsv.py --csv=results-bz619858.csv --avmname baseline --avm2name bz619858 which processes a csv file generated with the above --avmname/--avm2name values. I have not yet seriously tested processing a .csv generated without an --avm2 option or a .csv that does not explicitly provide nicknames via --avmname/--avm2name. (I do know that the -m option does not work at all; I didn't really think about it until now.)

Comment on attachment 501180 [details] [diff] [review] refactor 1: cut-and-pasting runtestsCommon.py Looks good to me. We could also move runtestCommon.py to ../utils (named something like runtestPerfBase.py?) so that it doesn't clutter the performance dir, but i'm ok with it either way.

(In reply to comment #1) > Q for Chris: what is "the real" Pythonic way to accomplish what I'm trying to > do in this patch? Is this an abuse of import? (Is "." always on the import > path?) Yes, "." will always be on the import path.

(it would be nice to land this, or something like it, but i'm not going to do it in near term.)

(I'm the only person who cared about this. Well, maybe Chris too. We have a Performance Team now, but I don't think they will be reusing this existing performance eval infrastructure.)