Specialized row decoders per type - #48

Draft

mzabani wants to merge 22 commits into

cps-parser-fewer-bytestringsfrom

specialized-row-decoders-per-type

Commits on Aug 23, 2026

Try a new kind of very specialized parser for small types
mzabani
committed
525ea20
Post-rebase fixing
mzabani
committed
b4d1114
More specialized instances, more confirmation of benefits
mzabani
committed
05e3a9d
Specialized instance for UTCTime => clear benefits once again
mzabani
committed
0be5704
Tidy up a bit
mzabani
committed
68378ee
Slightly better understanding of inlining, clearer inlining boundaries
mzabani
committed
feabc78
Add `inlinedSingleFieldDecoder`
mzabani
committed
7030fd5
More comprehensive coverage of types in benchmarks, more specialized row decoders
mzabani
committed
3bf360b
Tests for non-specialized field decoders
mzabani
committed
53b339c
Tidy up GHC Core
mzabani
committed
d46a807
Overlapping Maybe instances do help with the inlined row decoder's performance
mzabani
committed
ef7ce8d
Very experimental change with `Maybe a` instances
mzabani
committed
2e28392
A separate field in FieldDecoder for what to decode NULL to
mzabani
committed
fbfa92b
Try to improve the code, but inlining got worse
mzabani
committed
08cab31
Some memory usage improvements
mzabani
committed
e42a7bc
Float decodesSqlNullTo outside and add strictness for better inlining
mzabani
committed
2b91148
Tidy up, a few more INLINE pragmas
mzabani
committed
cc75921
TODOs in the code
mzabani
committed
5c7f52f
Trying a specialized notConst method
mzabani
committed
2a2e436
Do the JSON types, but hpgsql-simple-compat will break
mzabani
committed
9c02dc1
Post-rebase fixes
mzabani
committed
c0c7003
Update TODO
mzabani
committed
4861e73

Navigation Menu

Specialized row decoders per type - #48

Specialized row decoders per type#48
mzabani wants to merge 22 commits into
cps-parser-fewer-bytestringsfrom
specialized-row-decoders-per-type

Commits on Aug 23, 2026

Try a new kind of very specialized parser for small types

Post-rebase fixing

More specialized instances, more confirmation of benefits

Specialized instance for UTCTime => clear benefits once again

Tidy up a bit

Slightly better understanding of inlining, clearer inlining boundaries

Add `inlinedSingleFieldDecoder`

More comprehensive coverage of types in benchmarks, more specialized row decoders

Tests for non-specialized field decoders

Tidy up GHC Core

Overlapping Maybe instances do help with the inlined row decoder's performance

Very experimental change with `Maybe a` instances

A separate field in FieldDecoder for what to decode NULL to

Try to improve the code, but inlining got worse

Some memory usage improvements

Float decodesSqlNullTo outside and add strictness for better inlining

Tidy up, a few more INLINE pragmas

TODOs in the code

Trying a specialized notConst method

Do the JSON types, but hpgsql-simple-compat will break

Post-rebase fixes

Update TODO