[parker89]

Builder deserialization currently goes through MethodProperty, even though Blackbird already creates an accessor that returns the setter's result. Keep that accessor and use it from setAndReturn and deserializeSetAndReturn.

Second in the stack: #366 → #364 → #367. #366 is merged, so this diff now has only the builder change.

A setter's non-null result becomes the next builder; a null result keeps the current builder. Primitive int, long and boolean arguments avoid boxing. Void setters, incompatible values or receivers, and setters Jackson leaves subject to normal access checks continue through reflection. Builder properties that databind assigns through set() (buffered before a property-based creator, or external type ids) are handled the same way, so the outcome doesn't depend on the order of properties in the JSON. Null handling and setter error reporting match MethodProperty, including JSON location. (Builder setter failures already lacked the location before this PR, since MethodProperty.setAndReturn has no parser either.)

Once the generated setter is called, a failure is never retried, so a setter never runs twice. Telling a failure of the generated accessor apart from the setter's own would mean guessing from the stack trace, so both are reported like a setter failure. #366 prevents the known accessor failures at setup. The trade-off is that an accessor failure those checks miss now fails deserialization instead of falling back to reflection. #367 does the same for regular bean setters.

Tests verify that mutable and copying builders use generated setters, compare edge cases and failures with vanilla Jackson, and check that real JVM access failures are reported without running the setter. All Blackbird tests pass on Java 8, 11, 17 and 25.

The new builderArrayDeser benchmark deserializes 1,000 beans per operation. On JDK 25, allocation fell from 249.5 to 209.5 KB/op, mostly from avoiding primitive boxing. These measurements predate the review changes that added receiver/type checks. Throughput varied too much on my laptop to quote a reliable improvement.

Benchmark commands (use a disposable checkout)

# Run from a clean checkout of this branch. The benchmarks POM is outside the
# reactor and its parent is pinned to 2.19.0, so point it at the snapshot that
# each step installs, and build the benchmarks from this branch both times.
set -euo pipefail
BASE=644e951f      # merge base with 2.x
HEAD=$(git rev-parse HEAD)
JMH_ARGS=(-f 2 -wi 5 -w 2s -i 5 -r 3s -prof gc)

build() {  # $1 = commit to install Blackbird from, $2 = output jar
  git checkout --no-overlay -q "$1" -- blackbird afterburner
  ./mvnw -B -q -ntp clean install -pl blackbird,afterburner -am -DskipTests
  git checkout --no-overlay -q "$HEAD" -- blackbird afterburner
  (cd blackbird/benchmarks \
    && sed -i.bak 's|<version>2.19.0</version>|<version>2.23.0-SNAPSHOT</version>|' pom.xml \
    && ../../mvnw -B -q -ntp clean package -DskipTests \
    && mv pom.xml.bak pom.xml)
  cp blackbird/benchmarks/target/benchmarks.jar "$2"
}

build "$BASE" /tmp/benchmarks-base.jar
build "$HEAD" /tmp/benchmarks-head.jar
java -jar /tmp/benchmarks-base.jar '\.Blackbird\.builderArrayDeser$' "${JMH_ARGS[@]}"
java -jar /tmp/benchmarks-head.jar '\.Blackbird\.builderArrayDeser$' "${JMH_ARGS[@]}"

CLA submitted by email.

[cowtowncoder]

@stevenschlansker WDYT? Obviously only matters pre-rewrite but could merge in 2.x / merge forward to 3.1, 3.2, if this makes sense.

[cowtowncoder]

A few issues found while reviewing; details inline.

Claude Code on behalf of @cowtowncoder

[parker89]

Thanks for the review! Fixing it grew the change quite a bit, so I split it into three stacked PRs, to be merged in order:

1. Blackbird: skip generated setters the accessor can't call #366: setup checks so Blackbird skips generated setters it can't call. These also fix existing problems with regular beans, so it stands on its own.
2. This PR: generated setters for builders.
3. Blackbird: avoid retrying failed bean setters #367: no more retries of failed regular setters, plus Blackbird ignores @JsonSetter(nulls = Nulls.SKIP) for properties of beans with a properties-based creator #363.

Each PR's diff includes the earlier ones until they're merged, so #366 is the place to start.

[cowtowncoder]

+1 for splitting into stacked PRs! I'll need to weave changes wrt #359 that goes in 3.x so this might slightly simplify that too.

[parker89]

Sounds good, thanks for merging #366 . I'll hold off on any further changes to this PR or #367 until you give me the go ahead