197113 - nsConverterInputStream shouldn't buffer incoming bytes

Status: NEW

(Core :: Internationalization, defect, P3)

Description

[Alec Flett]

Continuing in my quest to reduce excess buffer copying, I've noticed that both
implementations of nsIUnicharInputStream (the UTF8-only one in XPCOM, as well as
the generic converter in intl/uconv/src) both use byte buffers to buffer input
in order to do the conversion. There's really no need for it, as the input
stream should be buffered or backgrounded as appropriate, and converted on
demand when someone calls nsIUnicharInputStream::Read() or
nsIUnicharInputStream::ReadSegments()

(in fact they don't even need to buffer much of the unicode output, if the
consumers are only calling ReadSegments() but that's a whole other beast!)

Comment 1

[Alec Flett]

would be nice, but no time left in 1.4beta.

Comment 2

[Sylvestre Ledru [:Sylvestre]]

Moving to p3 because no activity for at least 1 year(s).
See https://github.com/mozilla/bug-handling/blob/master/policy/triage-bugzilla.md#how-do-you-triage for more information

Comment 3

[Henri Sivonen (:hsivonen)]

UTF8InputStream is gone, but nsConverterInputStream still copies uselessly.