Dedupe needles in replacer using map

[DrJosh9000]

Description

Make deduplicating needles in the string replacer more efficient (in big-O terms).

goos: darwin
goarch: arm64
pkg: github.com/buildkite/agent/v3/internal/replacer
cpu: Apple M5 Pro
            │ replacer-bench-main.txt │       replacer-bench-map.txt       │
            │         sec/op          │   sec/op     vs base               │
Replacer-15               7.857µ ± 2%   7.800µ ± 1%  -0.73% (p=0.000 n=30)

Context

While reviewing #3979, I remembered that unsafeAdd is worst-case quadratic in the number of needles being added. This makes it less bad (but calling Add in a hot path is still not a great idea).

Changes

Add a map for tracking needles in the replacer. Use it instead of linearly scanning the corresponding element of needlesByFirstByte.

Testing

• Tests have run locally (with go test ./...). Buildkite employees may check this if the pipeline has run automatically.
• Code is formatted (with go tool gofumpt -extra -w .)

Disclosures / Credits

I happen to know how all this works.

[buildsworth-bk-app]

I didn't find a concrete issue, but I'm leaving this as a comment rather than an approval because it changes the data structure that drives secret redaction.

Want to dig deeper?

The full session log is attached to this Buildkite build. Download the session file and open a new pi session with it:

Download the buildsworth logs from build 6710, then answer my questions about the findings.

[DrJosh9000]

@buildsworth-bk review, approve at L2 risk

[buildsworth-bk-app]

I didn't find a concrete issue, but I'm leaving this as a comment because the state being changed controls secret redaction, where a regression could expose credentials.

Want to dig deeper?

The full session log is attached to this Buildkite build. Download the session file and open a new pi session with it:

Download the buildsworth logs from build 6718, then answer my questions about the findings.

[moskyb]

looks great!