[bndikt]

Status

• Draft; depends on feat: add DIRAC grid submission support #538 and is not intended for merge before that PR.
• This branch currently includes the grid-submission implementation from feat: add DIRAC grid submission support #538. After feat: add DIRAC grid submission support #538 merges, this branch will be rebased onto upstream/master, leaving only the input-resolution changes in the diff.
• For early inspection, the resolver-specific commits begin with 8037ea7 (“Consolidate runner input resolution”).

Changes

• Add a shared input_resolver.py for physical input paths, replacing the normal runner’s local-only directory discovery and the grid backend’s separate XRootD discovery logic.
• Resolve local paths, mounted CERN EOS paths, and XRootD URLs. Directory discovery is immediate and non-recursive; results contain sorted, deduplicated .root files.
• Translate /eos/experiment/fcc/... and /eos/user/... paths automatically. Other EOS namespaces require an explicit XRootD URL.
• For the normal runner, select inputs in this order:
  • CLI input, which overrides input settings in the analysis script;
  • a per-sample source: input-dir, input-files, or input-file-list;
  • global Analysis.input_dir;
  • campaign discovery as the fallback.
• Require CLI input mechanisms to be mutually exclusive, and require each sample to select at most one input source. A per-sample input source does not require an otherwise unused global input_dir.
• Support grid physical inputs through --input or --input-file-list, or through input settings in the analysis script. Grid physical inputs must be accessible through EOS or XRootD; local filesystem paths cannot be accessed by grid workers.
• Support grid LFNs through --input-lfn-list, Analysis.input_lfn_list, or a sample’s input-lfn-list setting. LFN parsing and DIRAC staging remain grid-specific.
• Keep physical inputs and LFNs as separate modes; they cannot be mixed in one grid submission. Campaign discovery remains available to the normal runner but is not yet supported for grid submission.
• Remove the deprecated --files-list option and update affected examples to use --input.
• Preserve compatibility with legacy RDFanalysis and build_graph scripts that do not use the files_list argument.
• Fix integration regressions for data-source Analysis scripts that do not define samples.

Testing and documentation

• Add tests for shared input resolution, input-source selection, and grid input planning.
• Update the related man pages.

[bndikt]

rework done - ready for review once #538 was merged

[jmcarcell]

Alternatively you could use stacked prs for dependent branches (allows reviewing each independently) https://docs.github.com/en/pull-requests/get-started/about-stacked-prs

[bndikt]

Alternatively you could use stacked prs for dependent branches (allows reviewing each independently) https://docs.github.com/en/pull-requests/get-started/about-stacked-prs

Good point. But I suppose I would need to create a temporary branch for this to have the desired effect of showing only the diffs with respect to its new base (i.e. #538) (I cannot choose a branch of my fork to base this one on). If you agree with that, I'll give it a shot.