As you mention streams are existing since the begining of times and it's incredible that this does not exist on a web platform, but apparently the subject is still not so easy, node changed its stream quite a lot of time.

I probably will not be able to answer back in the coming days but your judgement is tough, this API is the concatenation of thoughts of the Overlap thread. Indeed it must clarify the writable/readable aspects, as well as the flow control/congestion, moreover than, despite of what you are saying, the API does support multiple consumers, and I/O, it's close to node streams.

Of course you must "conflate" text and binary, you can not spend your time converting to text, to ArrayBuffer, to ArrayBufferView, to Blob, the user API knows what it is streaming.



Le 30/10/2013 19:04, Domenic Denicola a écrit :
From: Arthur Barstow []

If you have any comments or concerns about this proposal, please reply to
this e-mail by November 3 at the latest.
I have some concerns about this proposal, and do not think it is solving the 
problem at hand in an appropriate fashion. I believe it does not provide the 
appropriate primitives the web platform needs for a streams API. Most 
seriously, I think it has ignored the majority of lessons learned from existing 
JavaScript streaming APIs.

Here are specific critiques, in ascending order of importance.

- It has a read(n) interface, which is not valuable [1] but constraints the API 
in several awkward ways.

- It assumes MIME types are at all relevant to a streaming data abstraction, 
when this is not at all the case.

- In general, is far too backward-looking in attempting to integrate things 
like blobs or object URLs into what is supposed to be a forward-looking 
primitive for the future of the extensible web. Replacing these various 
disparate concepts is what developers want from streams [2].

- It conflates text streams and binary data. As outlined in previous messages 
[1], what type of data the stream contains *must* be an *immutable* property of 
the stream. In contrast, the proposed API actively encourage mixing multiple 
data types within a stream via readType, readEncoding, and the overloaded write 

- It conflates readable and writable streams, which prevents a whole class of 
abstractions and use cases like: read-only file streams; write-only HTTP 
requests; and duplex streams which read to one channel and write to another 
(e.g. a websocket, where writing pushes data to the server and reading reads 
data from the server---not the written data, but data that the server writes to 
you). Indeed, the only use case this proposal supports is transform streams, 
which take data in one end and output new data on the other end.

- It provides no mechanism for backpressure signaling to readable stream data 
sources or from writable stream sinks. As we have heard previously, any stream 
API that does not treat backpressure as a primary issue is not a stream API at 
all. [3]

- More generally, it does not operate at the correct level of abstraction, 
which should be close to the I/O primitives streams are meant to expose. This 
is evident in the general lack of APIs  for handing interaction with and 
signaling to underlying I/O sources or sinks.

- It's pipe mechanism is poorly thought out, and does not build on top of the 
existing primitives; indeed, it seems to install some kind of mutex lock that 
prevents the other primitives from being used until the pipe is complete. The 
primitives do not support multiple consumers, so the pipe mechanism handles 
that case in an ad-hoc way. It is not composable into chains in the fashion of 
traditional stream piping. Its lack of backpressure support prevents expression 
of key use cases such as piping a fast data connection (e.g. disk) to a slow 
data connection (e.g. push to a slow mobile device); piping a slow connection 
to a fast one; piping through encoders/compressers/decoders/decompressors; and 
so on. In general, it appears to take no inspiration from prior art, which is a 
shame since existing stream implementations all agree on how pipe should work.

In light of these critiques, I feel that this API is not worth pursuing and 
should not proceed to Working Draft status. If we are to bring streaming data 
to the web platform, we should instead do it correctly, learning the lessons of 
the JavaScript stream APIs that came before us, and provide a powerful 
primitive that gives developers what they have asked for and serves as 
something we can layer the web's many streaming I/O interfaces on top of.

I have concrete suggestions as to what such an API could look like—and, more 
importantly, how its semantics would significantly differ from this one—which I 
hope to flesh out and share more broadly by the end of this weekend. However, 
since the call for comments phase has commenced, I thought it important to 
voice these objections as soon as possible.



Peersm :
node-Tor :
GitHub :

Reply via email to