Closed dschwoerer closed 9 months ago
Merging #250 (5c82b92) into master (afb5ba6) will increase coverage by
0.00%
. The diff coverage is67.44%
.
@@ Coverage Diff @@
## master #250 +/- ##
=======================================
Coverage 68.99% 68.99%
=======================================
Files 15 15
Lines 3209 3248 +39
Branches 780 794 +14
=======================================
+ Hits 2214 2241 +27
- Misses 733 743 +10
- Partials 262 264 +2
Impacted Files | Coverage Δ | |
---|---|---|
xbout/load.py | 69.40% <67.44%> (-0.02%) |
:arrow_down: |
:mega: We’re building smart automated test selection to slash your CI/CD build times. Learn more
@dschwoerer what is the use case motivating this change?
I was certain I had replied already. Sorry for not doing so.
The motivation is to allow something like BOUT.dmp.*.nc
with having old, stale dump files.
I think the feature is nice, but the case I'd think of is where the run directories are not named in the expected way (e.g.
run0/
,run1/
, etc.). For that this PR doesn't help (yet!). It should be possible to getrun_restart_from
as well asrun_id
and then (assuming that the runs are a set of consecutive restarts, which presumably they usually will be if they are being concatenated in the time dimension) sortrun_ids
,run_restart_from_list
andpaths_sorted
so thatrun_restart_from_list[i] == run_ids[i-1]
.
That would also be nice but it is out of scope for this PR
Also it would be good to test the new functionality in
test_load
. Not sure how complicated that would be though, I guess you would need to add some functionality to generate somerun_id
values.
I didn't get the time to figure out how the tests work. And the tests being horribly slow does not help :-(
The motivation is to allow something like
BOUT.dmp.*.nc
with having old, stale dump files.
I'm not sure I'm quite following. This is already a nice improvement - that currently gives much nicer error messages, and could maybe be expanded to add more features - but it looks like this still opens all the files that are matched by the glob pattern. If I understand your comment, you want to handle something like running a simulation on 8 procs (the old stale simulation), then running a new one on 4 procs (the simulation we now want to open), and open with just BOUT.dmp.*.nc
as the pattern. I agree this would be nice, but it seems like in this example e.g. BOUT.dmp.7.nc
would still be added to paths_grid
to be opened, although that is a 'stale file' - am I missing something?
I would have liked to add tests, but I doubt to get to that anytime soon :-(
Also provide better error in some cases.