joconnor22 / life Goto Github PK

If you run the tests like this, from the LIFE directory rather than the testing directory (the README is unclear about this):

$ git clone [email protected]:joconnor22/LIFE.git
Cloning into 'LIFE'...
remote: Enumerating objects: 144, done.
remote: Counting objects: 100% (70/70), done.
remote: Compressing objects: 100% (34/34), done.
remote: Total 144 (delta 45), reused 36 (delta 36), pack-reused 74
Receiving objects: 100% (144/144), 89.68 KiB | 553.00 KiB/s, done.
Resolving deltas: 100% (68/68), done.
$ cd LIFE
$ testing/store-ref-data.sh 

Storing new * RefData!

cp: ../examples/*/params.h: No such file or directory
$ ls -a
*		.		..		.git		.gitignore

it deletes everything in the LIFE directory! This could be quite "irritating" if you happened to have uncommitted changes that could not be recovered from .git! (This just happened to me, and I luckily had a backup from yesterday, so could recover everything.)

The problem, as usual, is that bash is a terribly unsafe language. Adding some debug code, we see

$ testing/store-ref-data.sh 
+ for d in '../examples/*/'
+ casePath='../examples/*'
+ caseName='*'
+ printf '\nStoring new * RefData!\n\n'

Storing new * RefData!

+ echo 'rm -rf *'
rm -rf *
+ echo Exiting
Exiting
+ exit 0

Since '../examples/*/' doesn't match anything, since we are in the wrong directory, bash returns it unevaluated, and a lucky sequence of path manipulations converts this into rm -rf *. At least it doesn't delete anything outside LIFE! The (weird) behaviour of bash can be fixed by using

shopt -s nullglob

so that the empty wildcard expansion expands to a null string, rather than to itself. See here. The loop is then never run, and nothing is deleted.

It would also be good to have an initial check for being in the correct directory so that the user is given a good error message.

I will make a PR.

Currently, LIFE writes restart files overwriting the previous ones. If there is a failure during writing the restart file, the original restart file will be lost and the simulation will have to be started from the beginning. LIFE could instead write the new restart file to a temporary filename, and only when all restart files have been successfully written, rename them over the old ones.

On my machine, the LIFE test suite takes about 6 minutes to run, making it less likely that people will run them. I notice that the test cases run a lot of timesteps. If the purpose is to test for regressions, then probably only a few timesteps are needed. The resolution also doesn't need to be very high.

Would it make sense to reduce the number of timesteps, and potentially also the resolution of each test?

Dear developers，
It is my honor to write this letter！ I hope to investigate the effect of surface roughness on the deformation behavior of flexible plates in flow fields. However, due to a lack of understanding of the underlying logic of the project, I am not sure whether this open project can achieve my expected goals and whether there is complete validation of reference data and literature. I hope to receive a response from the developers, and I will be very grateful! Thank you!