On Wed, 14 May 2025 17:52:37 GMT, Chen Liang <li...@openjdk.org> wrote:
>> src/java.base/share/classes/java/io/Reader.java line 500:
>>
>>> 498: * @since 25
>>> 499: */
>>> 500: public String readAllAsString() throws IOException {
>>
>> Still thinking that declaring `CharSequence` instead of `String` would be
>> beneficial. In case a `Reader` implementation reads from I/O, the
>> implementation could so return a block of native memory without turning it
>> into a Java `String` just for sake of fulfilling the API. In case of servers
>> for example, the information often is passed-through from one I/O source to
>> another I/O sink *unmodified*, which means, the `String` then is turned into
>> *a new* native memory block again; double native memory used *plus* on-heap
>> memory used, without any benefit. Can we please provide I/O APIs which do
>> *not* enforce duplicate copying to and from `String` in such scenarios? 😃
>
> I think this is just a convenience API. A CharSequence-returning API would
> suit better if it supports arbitrarily long lengths (assuming such a length
> is in the range of int), while String has implementation limits on lengths
> and is allowed to fail for this API with OOME. If we make this a
> CharSequence-returning API that fails with OOME for String implementation
> limits, I don't think this will please anybody.
As being a server developer for decades I need to say that I would love to have
this not for convenience but for the performance reasons explained above, and
that `int` length is perfect to speed up 99,9% of all cases. So we should not
forget about `CharSequence` hastily.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/24728#discussion_r2089463475
