Environment variables with spaces break execution on clusters when using jobscript

[pcm32]

Snakemake version

6.2.1

Describe the bug

When the workflow makes use of environment variables (by declaring them in envvars and then using them through os.environ), the workflow execution fails on batch schedulers because the jobscript gets the variable split by the space (it needs quoting).

Logs

Look at COVARIATE below

/nfs/my-path/.snakemake/tmp.fjbaxuov/snakejob.retrieve_data.3.sh: line 3: CACHE_PATH=/my-path/cache ACCESSIONS=E-MTAB-4395,E-MTAB-4342,E-MTAB-4128,E-MTAB-3826,E-MTAB-3173,E-MTAB-964,E-GEOD-62778,E-GEOD-54272 BATCH=study NEW_ACCESSION=E-CORN-1 COVARIATE=organism part COVARIATE_TYPE=characteristic: No such file or directory

Minimal example

Run any workflow that uses an env variable that has a space in the value in a cluster environment. On a local execution (no cluster), there is no issue.

Additional context

I suspect that the issue is here: https://github.com/snakemake/snakemake/blob/main/snakemake/executors/__init__.py#L795
although changing that on my side doesn't resolve the problem entirely, but I'm sure that the unquoted values are not helping. Unfortunately debugging this is difficult since the jobscript (snakejob.retrieve_data.3.sh) is deleted when snakemake exists.

[pcm32]

I went into the code and saved the file separately to another location. Besides the space, the problem seems to be that all envvars get surrounded in single quotes, acting as a single entity. This is resolved by setting https://github.com/snakemake/snakemake/blob/main/snakemake/executors/__init__.py#L801 to False. After that all seems to work. Will PR.