Changed behaviour in handling template variables passed via cli

[jamesgrimmett]

Description

I think that there has been a change in behaviour in handling template variables passed to workflows via the -S cli option in v8.2.5.
In v8.2.3, a template variable passed via -S would override that variable set in optional config files. In v8.2.5, it looks like this behaviour is reversed, and the setting in optional config will override the setting passed to -S.

Reproducible Example

Example test suite containing;

test_workflow
├── flow.cylc
├── rose-suite.conf
├── opt
    └── rose-suite-common.conf

[flow.cylc]

#!jinja2
[scheduler]
   allow implicit tasks=True

[scheduling]
  initial cycle point = {{ START_CYCLE }}
  final cycle point = {{ END_CYCLE }}
  [[ graph]]
    R1 = cold_start
    T00 = """
      cold_start => a
    a => b
    """

[opt/rose-suite-common.conf]

[template variables]
START_CYCLE="20240101T00"
END_CYCLE="20240102T00"

Variables passed using -S are ignored, and the workflow runs with the settings from opt/rose-suite-common.conf;

$ cylc --version
8.2.5

$ cylc vip -O common -S 'START_CYCLE="20230102"' -S 'END_CYCLE="20230102"' .

$ ls ~/cylc-run/test_workflow/run1/log/job/
20240101T0000Z  20240102T0000Z

Expected Behaviour

The setting passed via -S overrides other config

$ cylc --version
8.2.3

$ cylc vip -O common -S 'START_CYCLE="20230102"' -S 'END_CYCLE="20230102"' .

$ ls ~/cylc-run/test_workflow/run1/log/job/
20230102T0000Z

I seems that I get the behaviour that I'm looking for if I use -s instead of -S, so I could potentially use that as a workaround, but I don't have a good understanding of the difference between these two options. I note that -s will currently not allow variables to be updated on reload (#5965 / #6045), are there any other significant differences between these two options (i.e., -s vs -S)?

$ cylc --version
8.2.5

$ cylc vip -O common -s 'START_CYCLE="20230102"' -s 'END_CYCLE="20230102"' .

$ ls ~/cylc-run/test_workflow/run1/log/job/
20230102T0000Z

[hjoliver]

.. I get the behaviour that I'm looking for if I use -s instead of -S, ... but I don't have a good understanding of the difference between these two options.

The -S option is injected into Cylc by the cylc-rose plugin, so I'll leave this one to the UK Rose team overnight...
They'll probably want to move this issue to the cylc-rose repository.

[wxtim]

-s is a native Cylc feature: As such it ought to be safe as a work-around.

I will examine this when I get into the office.

[oliver-sanders]

We recently changed this: cylc/cylc-rose#302

Shouldn't be related?

[wxtim]

We recently changed this: cylc/cylc-rose#302

Shouldn't be related?

I don't think* that should be causing this change.

* To a very high level of certainty.

[wxtim]

I can confirm that this

• This looks like a bug.
• It applies to both cylc install && cylc play and to cylc vip (i.e. it's not a side effect of something VIP is doing).
• It is happening with the most recent versions of Cylc and Cylc-Rose.

Results from multiple runs suggest that this may be a change to Cylc.

===============
8.2.4+1.3.x
...............
CLI
===============
8.2.5+1.3.0
...............
opt_file
===============
8.2.5+1.3.2
...............
opt_file
===============
8.2.5+1.3.3
...............
opt_file
===============
8.2.5+1.3.x
...............
opt_file
===============
8.2.x+1.3.x
...............
opt_file
===============
master+1.3.x
...............
opt_file

Cause of the bug!

#5996

[wxtim]

Overview of solution

For Cylc 8.2.x:

#6059 - Cleaned all 3 config items on this branch.

For Cylc 8.3.0 and Cylc Rose 1.4.0 (master branches in both repos)

#6060 - remove all the cleaning code from Cylc-flow
cylc/cylc-rose#308 - added the option removal to Cylc Rose.

[oliver-sanders]

Closed by cylc/cylc-rose#312