feat: Complete `exclude_newer` policy by jjerphan · Pull Request #4344 · mamba-org/mamba

jjerphan · 2026-07-01T10:55:15Z

Description

Continues #4228 by completing the exclude_newer policy: wiring config/CLI through to the solver, adding per-package overrides, and hardening duration/date parsing.

exclude_newer / exclude_newer_package: global cutoff plus per-package overrides (false to exempt, or a duration/date/timestamp). Available via RC, CONDA_/MAMBA_ env vars, and CLI (--exclude-newer, --exclude-newer-package as a JSON/YAML dictionary).
ExcludeNewerPolicy: resolved cutoffs only (global / per_package); raw config stays on Context and is resolved at solve time via resolve_exclude_newer_policy.
Parsing: compact durations (7d, 3d12h), ISO 8601 durations (P7D), date-only / datetime values; uses std::chrono::parse when available, otherwise Howard Hinnant's date (needed on macOS). Unparseable values throw mamba_error.
Build/runtime: howardhinnant_date for macOS static/dev builds; declare msvcp140_atomic_wait.dll for Windows micromamba.

Background / cross-ecosystem tracking: conda/conda#15759

Type of Change

Bug fix
Feature / enhancement
CI / Documentation
Maintenance

Checklist

My code follows the general style and conventions of the codebase, ensuring consistency
I have performed a self-review of my code
I have commented my code, particularly in hard-to-understand areas
My changes generate no new warnings
I have run pre-commit run --all locally in the source folder and confirmed that there are no linter errors.
I have added tests that prove my fix is effective or that my feature works
New and existing tests pass locally with my changes

Test plan

Unit tests: test_exclude_newer.cpp, database exclude-newer sections, libmambapy exclude_newer_timestamp tests
Config: {CONDA,MAMBA}_EXCLUDE_NEWER{,_PACKAGE} and micromamba install --exclude-newer / --exclude-newer-package … --print-config-only
Smoke: install with --exclude-newer 7d and a per-package exemption; confirm newer builds are filtered and exempted packages remain
Confirm macOS static build picks up howardhinnant_date-static and Windows DLL allowlist includes msvcp140_atomic_wait.dll

Klaim

I'll make another review pass later, I didnt reach half of the changes yet.

Klaim · 2026-07-23T11:22:17Z

+     * Returns ``std::nullopt`` when parsing fails or trailing characters remain.
+     */
+    template <typename T>
+    [[nodiscard]] auto parse_chrono(std::string_view value, const char* fmt) -> std::optional<T>


Just to be clear: this is necessary only because MacOS's build toolchain doesnt implement std::chrono::parse yet?
If so, could you add a TODO about this in the documentation comments to replace it when std::chrono::parse becomes available everywhere?

Added a TODO to drop the Howard Hinnant fallback once std::chrono::parse is ubiquitous on supported platforms.

Klaim · 2026-07-23T11:31:29Z

+         *
+         * See Configurable ``exclude_newer_package``.
+         */
+        std::map<std::string, std::string> exclude_newer_package;


In the end this is never used as a registry as resolve_exclude_newer_package_cutoffs will just go through elements one by one. I think it would be better if it was a vector of struct, or pair if that helps with CLI11 ?

Switched Context::exclude_newer_package to std::vector<std::pair<std::string, std::string>>; added YAML convert + dump_json so CLI/env JSON still round-trips as an object.

Klaim · 2026-07-23T11:42:18Z

+namespace mamba
+{
+    /**
+     * Resolved per-package ``exclude_newer`` cutoffs.


kind of side question: why are you using double anti-quote instead of just normal anti-quote for code parts?
If my understanding is correct, markdown is automatically handled by our doxygen setup?

Updated to single backticks for Doxygen Markdown.

Klaim · 2026-07-23T11:45:34Z

+     *
+     * The public API exposes ``std::uint64_t`` seconds for compatibility with repodata
+     * timestamps. Internally, date and datetime values are parsed with
+     * ``std::chrono::parse`` when the standard library provides it, otherwise with


I would avoid letting the details of the implementation appear in the description of the interface of a function when it's not a description of the general algorithm being implemented, so that changing the implementation does not make that comment/description obsolete without us noticing.

Removed implementation details from resolve_exclude_newer_cutoff; kept semantic description only.

Klaim · 2026-07-23T11:50:17Z

+            const std::map<std::string, std::string>& exclude_newer_package
+        ) -> solver::libsolv::Database::Settings
+        {
+            const auto now = static_cast<std::uint64_t>(std::time(nullptr));


Why are we not using chrono to get this timepoint value?

Reference time now uses std::chrono::system_clock.

(I thought that libc++ did not implemented this part, which is not the case.)

Klaim · 2026-07-23T11:53:08Z

+            std::string message;
+            if (package_name.empty())
+            {
+                message = "Could not parse exclude_newer value '" + std::string(value) + "'; "


Consider using fmt instead of string concatenation for generating these messages.

Parse errors now use fmt::format.

Klaim · 2026-07-23T12:38:04Z

+            static const std::regex compact_duration{ R"(^(\d+[yMwdhms])+$)" };
+            static const std::regex compact_segment{ R"((\d+)([yMwdhms]))" };
+
+            constexpr std::uint64_t seconds_per_minute = 60;


This is duplicated from the previous function.
Also I don't think "second per year/month" is correct (doesnt it changes depending on the year?), these values would be better deduced through chrono?

Deduplicated unit constants; year/month use std::chrono::years/months via duration_cast. Tests updated accordingly.

Klaim · 2026-07-23T12:41:28Z

+                    return std::nullopt;
+                }
+
+                std::uint64_t multiplier = 0;


Why not just use the facilities provided by std::chrono instead of doing it manually?

Compact parser now accumulates std::chrono durations instead of manual multipliers.

Apply clang-format to ternary expression in database.cpp and a pre-existing line-break issue in helpers.cpp.