sonatard / gqlgen / 39383006899
0%
master: 0%

LAST BUILD BRANCH: exec-behavior-tests
DEFAULT BRANCH: master
Ran
Jobs 1
Files 0
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 0.0%. Remained the same
39383006899

Pull #42

github

sonatard 
graphql/executor: make BenchmarkRequest B/op independent of b.N

BenchmarkRequest's B/op wobbled by about 1% between runs of the same code
(1787 to 1809 B/op over -count 8), which made the zero-tolerance
benchmark gate fail at random on unrelated PRs.

The noise was not GC or sync.Pool timing: with GOGC=off it remained,
and with a fixed -benchtime Nx it disappeared, while B/op changed
deterministically with N (1768 at 1000x, 1809 at 30000x). Two things
in the test fixtures made the bytes per op depend on b.N:

- testexecutor used graphql.ExecutableSchemaMock, whose Exec appends
  every call to a slice nothing reads. That slice grows with b.N, so
  its reallocations cost a different amount per op depending on where
  b.N falls between two of them. testexecutor now uses a small
  ExecutableSchema that only calls the functions it is given.
- The first measured iteration missed the query cache and parsed and
  validated the query, a one-off cost that b.N divided. The benchmark
  now serves one request before the loop, as
  BenchmarkCreateOperationContext already does for its cached case.

The executor itself did not contribute. With both changes B/op is 1721
in 15 of 16 runs and 1722 in the other, and stays at 1720 to 1722 from
100x to 100000x.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Pull Request #42: graphql/executor: make BenchmarkRequest B/op independent of b.N

0 of 0 relevant lines covered (NaN%)

0.0 hits per line

Jobs
ID Job ID Ran Files Coverage
1 39383006899.1 0
0.0
GitHub Action Run
Source Files on build 39383006899
Detailed source file information is not available for this build.