Try to fix CONFLICT on pop

[ecarreras]

• Restore dirty working-tree changes after applying a patch.
• Capture git stash pop failures without letting Fabric abort before cleanup.
• Detect unmerged index entries explicitly and reset a conflicting pop while preserving the stash.
• Keep compatibility with the current PatchApplier.apply(..., sudo_user=...) API.

Validation

• Python 2.7: pytest -q — 17 passed.

[guilleJB]

Could generate a git diff file before to stash dirty code and delete it when all runs well?

[ecarreras]

Could generate a git diff file before to stash dirty code and delete it when all runs well?

Why? the stash doesn't get deleted if we do a git reset --merge

[guilleJB]

Why? the stash doesn't get deleted if we do a git reset --merge

Sorry, I'm read git merge --reset not git reset --merge

[giscebot]

Requesting changes because the conflict detector is incorrect for the normal case and the actual conflict path is not reliably handled. I reviewed the full PR discussion and diff, traced Fabric command behavior, checked the current branch state, ran git diff --check, and reproduced the detector after a successful git stash pop in a temporary repository. The PR also has no CI checks and currently conflicts with master.

[ecarreras]

@giscebot please fix this with

[giscebot]

Fixed and pushed in 04f7e5f.

• corrected successful/conflicting git stash pop handling;
• added regression coverage for both paths;
• merged current master and resolved the PR conflict;
• Python 2.7 suite: 17 passed.

The PR is now mergeable (CLEAN). There are currently no configured GitHub checks on this PR.

[ecarreras]

@giscebot tingues en compte el mode "local"

[giscebot]

Sí, revisat tenint en compte el mode --local.

En local aquest cas no pot arribar a stash pop: apply_pr/local.py::validate_repository() rebutja el checkout si git status --porcelain no és buit, abans de registrar ni aplicar el desplegament. L’invariant ja està cobert per test_rejects_dirty_checkout_before_registering_deployment.

Per tant, la gestió de conflictes de restauració d’aquesta PR ha de continuar limitada al backend Fabric/remot (PatchApplier); afegir un stash implícit al mode local canviaria deliberadament el seu contracte de seguretat. Els conflictes de git am locals ja segueixen el seu flux propi (--abort amb auto_exit o resolució manual). No veig cap canvi addicional necessari a la PR pel mode local.