perf: remove redundant allocations & computations from generated code

Summary

• fix: decode field name in error from utf8
• fix: use format_ident instead of trim_start_matches
• fix: context used almost always to ignore context
• perf: remove redundant allocations & computations from generated code

Ensure that

• New code is covered by tests
• API is documented
• Changelog is up to date
• Version is bumped in the appropriate Cargo.toml files

marked this merge request as draft

assigned to @e-ivkov

added 1 commit

• 854dedc7 - fix: context used almost always to ignore context

Compare with previous version

This is what I get now when running benchmarks defined in !423 (merged)

custom_encode #0: 4.1183306s (41.183µs)
custom_encode #1: 4.092303s (40.923µs)
custom_encode #2: 4.1117257s (41.117µs)
serde_encode #0: 4.784078s (47.84µs)
serde_encode #1: 4.7692769s (47.692µs)
serde_encode #2: 4.8269951s (48.269µs)
custom_decode #0: 14.156257s (141.562µs)
custom_decode #1: 14.158553s (141.585µs)
custom_decode #2: 14.1795236s (141.795µs)
serde_decode #0: 18.828379s (188.283µs)
serde_decode #1: 18.8760486s (188.76µs)
serde_decode #2: 18.8533508s (188.533µs)

BTW in debug builds we also outperform serde

mentioned in merge request !427 (merged)

added 10 commits

• 854dedc7...7401f597 - 6 commits from branch master
• 1b361ad2 - perf: remove redundant allocations & computations from generated code
• 466509e0 - fix: context used almost always to ignore context
• a5d1ab85 - fix: use format_ident instead of trim_start_matches
• 9d6fb798 - fix: decode field name in error from utf8

Compare with previous version

resolved all threads

marked this merge request as ready

marked the checklist item New code is covered by tests as completed

requested review from @gmoshkin

approved this merge request

merged