Incorrect cache loaded if default not found

[2e0byo]

With the following workflow definition (excerpted):

- run: echo '-G dev -G ci' > deps.groups
- name: Install PDM
  uses: pdm-project/setup-pdm@v3
  with:
    python-version: "3.10"
    cache: true
    # These files will be hashed to generate the cache key
    cache-dependency-path: |
      ./pdm.lock
      ./deps.groups
- run: pdm sync $(cat deps.groups)

I would expect the behaviour to be:

First run

• pdm looks for a cache by hashing the contents of pdm.lock and deps.groups
• pdm doesn't find anything, so no cache is pulled
• pdm creates a cache with the deps installed in this run

Subsequent runs:

• pdm looks for a cache by hashing pdm.lock and deps.groups
• pdm finds the previous cache
• installs are quick, since the cache is populated

In actual fact the behaviour is:

First run

• pdm pulls another, unrelated pdm cache (with 7GB of unwanted deps)
• pdm installs the desired groups
• pdm creates a hash under the correct key, with 7GB of unwanted deps + new deps

Subsequent runs:

• pdm pulls the cache made in the first run
• installs take forever, because the cache is polluted

Presumably there's some fallback logic in cache.ts. I intend to PR a fix for this.

This breaks workflows where pdm is being used to handle multiple sets of deps (e.g. a tiny install for linting tools vs. everything for unit tests).

[2e0byo]

I have since worked out why this is the default behaviour: you want to pre-seed the cache when a dependency changes. I'll have a think and see if I can write a PR which satisfies both requirements.