Improve render performance

[kschweppe]

I noticed renderLines is causing a significant performance bottleneck when rendering large tables. This can easily be addressed by using Builder instead of Text.

Another small improvement is to construct Line directly with the correct indent to avoid unnecessary allocations caused by calling addIndent, which is called quite often.

Before:

COST CENTRE       MODULE                               SRC                                                             %time %alloc

renderLines       Web.Atomic.Render                    src/Web/Atomic/Render.hs:(177,1)-(187,30)                        80.2   98.8
parseEntityValues Database.Persist.Sql.Util            Database/Persist/Sql/Util.hs:(168,1)-(194,51)                     5.1    0.5
rawSelectSource   Database.Esqueleto.Internal.Internal src/Database/Esqueleto/Internal/Internal.hs:(3064,1)-(3080,40)    2.8    0.2
elementLines      Web.Atomic.Render                    src/Web/Atomic/Render.hs:(54,1)-(78,27)                           1.2    0.1

After:

COST CENTRE       MODULE                               SRC                                                             %time %alloc

parseEntityValues Database.Persist.Sql.Util            Database/Persist/Sql/Util.hs:(168,1)-(194,51)                    37.6   44.6
rawSelectSource   Database.Esqueleto.Internal.Internal src/Database/Esqueleto/Internal/Internal.hs:(3064,1)-(3080,40)   18.9   21.5
unsafeEff         Effectful.Internal.Monad             src/Effectful/Internal/Monad.hs:150:1-29                          4.1    0.8
elementLines      Web.Atomic.Render                    src/Web/Atomic/Render.hs:(60,1)-(82,79)                           3.7    4.0
column            Database.Sqlite                      Database/Sqlite.hs:(546,1)-(561,36)                               2.9    0.4
text              HTMLEntities.Text                    library/HTMLEntities/Text.hs:(24,1)-(25,21)                       2.5    3.5
fuse              Data.Conduit.Internal.Conduit        src/Data/Conduit/Internal/Conduit.hs:769:1-12                     2.1    1.3
stepConn          Database.Sqlite                      Database/Sqlite.hs:(331,1)-(336,46)                               1.8    0.0
renderLines       Web.Atomic.Render                    src/Web/Atomic/Render.hs:(178,1)-(188,44)                         1.6    2.0
tcol              Web.Hyperbole.View.Tag               src/Web/Hyperbole/View/Tag.hs:(224,1)-(225,69)                    1.1    0.3
char              HTMLEntities.Text                    library/HTMLEntities/Text.hs:(9,1)-(16,25)                        1.0    1.7
nodesLines        Web.Atomic.Render                    src/Web/Atomic/Render.hs:50:1-48                                  0.6    2.0

[seanhess]

This is awesome, thanks!

[seanhess]

@kschweppe could you update RenderSpec to match the way the new rendering works? It's currently calling addIndent in a couple tests.

Also main is now passing CI if you want to merge

[kschweppe]

Ah I missed the test. I just added addIndent again, or do you prefer to drop it and update the tests? Also the indent argument works different now, previously htmlLines 0 would mean no indentation, and now it always increments with a fixed step 2. Is this okay? Otherwise we could add an additional argument to have the same meaning together with the new implementation

elementLines :: Int -> Int -> Element -> [Line]
elementLines indent depth elm = ...

[seanhess]

We can remove addIndent, I just want to make sure the output is indenting correctly, so the tests would need to be updated.

[seanhess]

Also the indent argument works different now, previously htmlLines 0 would mean no indentation, and now it always increments with a fixed step 2. Is this okay? Otherwise we could add an additional argument to have the same meaning together with the new implementation

I don't mind if the argument meaning changes, as long as the output is properly indented.

[kschweppe]

Okay, should be good now.

Btw, I needed to add build-tool-depends: skeletest:skeletest-preprocessor to the cabal file, otherwise I would get ghc-9.10.3: Haskell pre-processor: could not execute: skeletest-preprocessor.

[seanhess]

Can't wait to get this into a release! Nice work!