fs: add WriteStream.prototype.flush

[himself65]

fix #28513

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• commit message follows commit guidelines

[bnoordhuis]

I appreciate that you took this on but this isn't quite what I had in mind. I tried to explain how it should work but don't hesitate to ask if anything is still unclear.

[himself65]

WIP

[himself65]

I'm trying to refactor the buffer poll to the task poll, which can solve if write chunk or do fsync

[addaleax]

@himself65 I’m not sure, but maybe #28520 can provide some inspiration? The code ultimately attempts to solve a very similar problem: Integrate flushing operations (and maybe this function here should also be called flush(), because that may be easier to understand than the POSIX function name) into a stream of to-be-written data. The basic idea there is to use special 0-sized buffers (in this case a single, global 0-sized buffer should suffice), which are then written using .write() and intercepted again in ._write().

[himself65]

so what is the case of writev?

example:

ws.cork()
ws.write('string')
ws.flush()  // or called ws.fsync()
ws.write('string')
ws.uncork()

in the current will call _writev

[addaleax]

@himself65 That’s indeed trickier… it might be okay to perform the fsync operation after all other buffers passed to the _writev() call? Otherwise the logic here would get pretty complex…

[addaleax]

Fwiw, the naming is unfortunate here, but please be aware that the stream _flush() method does something very different than what the idea for fsync() calls is; it’s called just after the last chunk of data has been written, and doesn’t have anything to do with actually flushing data to the underlying resource (it’s not a great name, definitely).

[jasnell]

@himself65 ... still want to do this?

Converting the PR to a draft given that it's still a work-in-progress