[Hamdy]

Checklist:

• I have updated the necessary documentation
• I have signed off all my commits as required by DCO
• My build is green

---

Summary

Each field's :if/:unless lookup was meant to be computed once, but it ran again for every object in every render.

Field#if_callable and #unless_callable cache their result with @ivar ||= callable_from(...). When a field has no condition, which is the common case, callable_from returns false. ||= never caches false, so both lookups ran again for every field of every object, and each one read Blueprinter.configuration.

The fix swaps ||= for a defined? guard, so a false result is cached too. It's two methods in lib/blueprinter/field.rb and adds no allocations. The result is about 40% less time per rendered object.

Benchmarks

I measured main against main plus this fix, using the benchmark suite from #614 (split out of this PR per review). Both ran on the same machine with Ruby 3.3.9, alternating over 5 rounds each. The numbers are medians.

Iterations per second (higher is better)

Benchmark	main	This PR	Change
Basic IPS	43,969	63,276	+43.9%
ActiveRecord IPS	46,684	66,107	+41.6%

Renders per second (higher is better)

Shape	main	This PR	Change
1 object × 6 fields	184,352	272,124	+47.6%
100 objects × 6 fields	2,304	3,786	+64.3%
1,000 objects × 6 fields	233.6	380.3	+62.8%
1,000 objects × 12 fields	118.2	193.6	+63.8%

Microseconds per object (lower is better)

Shape	main	This PR	Change
1 object × 6 fields	5.42	3.67	−32.3%
100 objects × 6 fields	4.34	2.64	−39.2%
1,000 objects × 6 fields	4.28	2.63	−38.6%
1,000 objects × 12 fields	8.46	5.17	−38.9%

Big-O render time, ms (lower is better)

Objects	Basic main	Basic PR	Change	ActiveRecord main	ActiveRecord PR	Change
100	0.276	0.221	−19.9%	0.401	0.295	−26.4%
1,000	2.641	1.943	−26.4%	3.714	2.736	−26.3%
10,000	27.228	20.583	−24.4%	36.258	26.262	−27.6%

Allocations per render (lower is better)

Shape	main	This PR	Change
1 object × 6 fields	10	10	0%
100 objects × 6 fields	108	108	0%
1,000 objects × 6 fields	1,008	1,008	0%
1,000 objects × 12 fields	1,008	1,008	0%

Field#callable_from calls per render (lower is better)

Objects	main	This PR	Change
1	12	12	0%
10	120	12	−90.0%
100	1,200	12	−99.0%
1,000	12,000	12	−99.9%

The last table explains the speedup. The lookup now runs once per field per render (6 fields × :if/:unless = 12) instead of once per field per object.

Behaviour change: please review

If a field has no :if/:unless of its own, it now captures the global config.if/config.unless on its first render. A global condition set after a blueprint has already rendered is ignored, where before it took effect. Setting it before the first render works exactly as before. The snapshots the global condition on first use spec pins this.

Tests

spec/units/field_spec.rb adds coverage for Field#skip?: field-level procs, symbols, global config, and precedence. Two examples fail on main and guard the regression: does not read global configuration again after the first call and snapshots the global condition on first use.

[lessthanjacob]

Thanks for pushing this up @Hamdy! The improvements look promising.

I left a few comments, but outside of that, I’d lean towards splitting out the additional benchmarks from the fix itself (given that it represents only a small fraction of the code here). We’ve had previous discussions around reworking the benchmark tooling previously, so I’d be happy to work with you on that separately from this change!