fix(conda): dedup repodata by archive identifier instead of URL

[jdx]

Summary

Fixes #9829 — mise use -g imagemagick (and other tools pulling adwaita-icon-theme transitively) fails with:

conda solve failed: encountered duplicate records for adwaita-icon-theme-40.1.1-ha770c72_1.tar.bz2

PR #8337 previously addressed this class of error by deduplicating records by URL in flatten_repodata, but that wasn't sufficient. rattler-solve's resolvo solver detects duplicates by DistArchiveIdentifier (name-version-build + archive type), not by URL — see rattler_solve-6.0.2/src/resolvo/mod.rs:414. When conda-forge serves the same archive under multiple URLs (distinct CDN paths / aliasing), URL-based dedup keeps both, so the solver still rejects them.

This PR switches the dedup key to r.identifier, the exact key rattler-solve uses, so colliding records can no longer reach the solver. .conda vs .tar.bz2 variants of the same package remain distinct (their archive_type differs), preserving the solver's existing archive-type preference logic.

Test plan

• cargo check -p mise compiles
• User-reported reproducer: mise use -g imagemagick on Linux64

---

Note

Medium Risk
Changes conda solver input deduplication to drop records by DistArchiveIdentifier, which can affect which package variants reach the solver and thus impact dependency resolution. Scope is small and covered by new regression tests, but it touches install/solve behavior.

Overview
Fixes conda solve failures caused by duplicate package records being returned under different URLs by deduplicating solver inputs on RepoDataRecord.identifier (archive identifier) instead of URL.

Refactors repodata flattening to reuse a new dedup_records_by_identifier helper, and adds unit tests to ensure identical identifiers across different URLs collapse while distinct .conda vs .tar.bz2 variants are preserved.

^{Reviewed by Cursor Bugbot for commit 01fc485. Bugbot is set up for automated code reviews on this repo. Configure here.}

[greptile-apps]

Greptile Summary

Fixes a conda solve failure where rattler-solve's resolvo solver rejects duplicate RepoDataRecords that share the same DistArchiveIdentifier (name-version-build + archive type) but were served under different CDN URLs. The previous URL-based dedup in flatten_repodata was insufficient for this case.

• Extracts a new dedup_records_by_identifier helper that filters on r.identifier (the exact key the solver uses), keeping the first-encountered record when two URLs resolve to the same archive.
• flatten_repodata now delegates to this helper, preserving .conda vs .tar.bz2 variants as distinct (their CondaArchiveType differs), so the solver's archive-type preference logic is unaffected.
• Two focused unit tests cover the regression case (same identifier, different URLs → collapse to 1) and the preservation case (.conda vs .tar.bz2 → keep both).

Confidence Score: 5/5

Safe to merge — the change is a single-file, narrowly scoped fix to conda repodata deduplication with no observable behavior change for non-duplicate inputs.

The fix correctly targets the exact key the solver uses to detect duplicates (r.identifier), the refactor into dedup_records_by_identifier preserves all existing behavior for the common case, and the two new unit tests cover both the regression scenario and the archive-type preservation invariant. No other code paths are affected.

No files require special attention.

Important Files Changed

Filename	Overview
src/backend/conda.rs	Replaces URL-based dedup with identifier-based dedup in flatten_repodata; extracts the logic into dedup_records_by_identifier; adds two regression unit tests. Change is minimal, targeted, and correct.

_{Reviews (4): Last reviewed commit: "style(conda): apply rustfmt to test addi..." | Re-trigger Greptile}

[gemini-code-assist]

Code Review

This pull request updates the deduplication logic in flatten_repodata to use archive identifiers instead of URLs, ensuring consistency with how rattler-solve detects duplicates. Feedback suggests optimizing the implementation by filtering references before cloning to improve performance and reduce unnecessary memory allocations.

[github-actions]

Hyperfine Performance

mise x -- echo

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
mise-2026.5.6 x -- echo	58.1 ± 11.2	31.1	81.3	1.00
mise x -- echo	59.9 ± 10.5	31.0	81.3	1.03 ± 0.27

mise env

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
mise-2026.5.6 env	56.3 ± 10.3	29.9	72.0	1.00
mise env	57.0 ± 10.3	26.6	76.7	1.01 ± 0.26

mise hook-env

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
mise-2026.5.6 hook-env	60.2 ± 11.8	28.3	79.1	1.00
mise hook-env	61.0 ± 10.5	32.3	84.5	1.01 ± 0.26

mise ls

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
mise-2026.5.6 ls	49.6 ± 8.8	26.0	68.2	1.00
mise ls	49.9 ± 8.6	21.2	66.8	1.01 ± 0.25

xtasks/test/perf

Command	mise-2026.5.6	mise	Variance
install (cached)	290ms	271ms	+7%
ls (cached)	190ms	189ms	+0%
bin-paths (cached)	206ms	207ms	+0%
task-ls (cached)	706ms	682ms	+3%

[thernstig]

Works great.