Avoid intermediate list when allocating bytestrings

[mchav]

These functions create an intermediate list. For large parquet files this is a lot of overhead. We can pre-allocate some space and create a bytestrin directly.

readByteString :: BS.ByteString -> IORef Int -> IO BS.ByteString
readByteString buf pos = do
    size <- readVarIntFromBuffer @Int buf pos
    BS.pack <$> replicateM size (readAndAdvance pos buf)

readByteString' :: BS.ByteString -> Int64 -> IO BS.ByteString
readByteString' buf size = BS.pack <$> mapM (`readSingleByte` buf) [0 .. (size - 1)]

[daikonradish]

DataHaskell has been accepted to Google Summer of Code 2026! This issue has been tagged with gsoc2026. We ask that this issue be reserved for those who are interested in applying to GSOC 2026.

If you are not applying to GSOC 2026, but you are still interested in this task, please approach us to discuss first. We're available on GitHub and on Discord.

Thank you!