FGasper / p5-Promise-XS / 24113385519

88%

push

github

Reimplement all method in XS (#12)

## Summary

  Implement `all()` in XS (C), replacing the previous pure-Perl implementation.

  ## Motivation

  The old Perl implementation chained `.then()` calls sequentially:

  ```perl
  $all_done = $all_done->then(sub {
      @results = @_;
      return $p;
  })->then(sub { (@results, [@_]) });
  ```

  This meant promise N was not awaited until promise N−1 had already resolved —
  contrary to Promise/A+ semantics. The new XS implementation attaches callbacks
  to all input promises independently, giving true parallel waiting.

  ## Changes

  - **`XS.xs`**: Add `XSPR_CALLBACK_ALL` callback type and `pxs_all_state_t` shared state. Each input promise gets its own callback; when all resolve the output promise is resolved with an
   array-ref per input. The first rejection short-circuits and propagates to the output promise; subsequent rejections are silently suppressed (their `rejection_should_warn` flag is
  cleared) so only one unhandled-rejection warning is emitted. Exception safety during initialisation is provided by `SAVEDESTRUCTOR_X` guards that free partially-constructed state on
  croak/die.
  - **`lib/Promise/XS/Promise.pm`**: Remove the now-superseded Perl `all()`.
  - **`t/unhandled_rejection.t`**: Add regression tests confirming that a caught `all()` rejection produces no warnings, and an uncaught one produces exactly one warning (from the output
  promise, not from each input).

  ## Behavioural notes for users

  | Aspect | Before | After |
  |--------|--------|-------|
  | Execution model | Sequential (chained `.then()`) | Parallel (independent callbacks) |
  | Return value format | `([v1…], [v2…], …)` | Same |
  | Unhandled rejection warnings | Could leak multiple warnings | Exactly one warning from output |
  | Performance | O(N) Perl call overhead | XS |

  The return-value format is unchanged. Code that accidentally relied on sequential execution (e.g. ordering side-effects through `all()`) sh... (continued)

63 of 63 new or added lines in 1 file covered. (100.0%)

728 of 830 relevant lines covered (87.71%)

548.86 hits per line