A javascript template reading a compressed response with Recv cannot
decode it without first holding the whole body. compress.NewDecoder
takes a Content-Encoding the HTTP engine decodes (gzip, deflate, br,
zstd, or identity) and decodes the chunks written to it, handing back
up to about 1 MB per call so a highly compressed chunk never has to be
held whole.
Go's decompressors pull from a reader while templates push chunks, so
each decoder runs its decompressor on a goroutine that pauses whenever
it needs input or has 1 MB to hand back. The goroutine exits when the
script's context ends, so an abandoned decoder does not leak.
Kamil Kamili
6c615b76df