On Thu, 24 Oct 2024 14:00:39 GMT, Jaikiran Pai <j...@openjdk.org> wrote:

>> readAllBytes reads to EOF so there isn't any benefit to caching. I think the 
>> main thing with this PR is whether canSeek is the right thing to use.
>
> Hello Alan, I didn't just mean the readAllBytes. The rest of the changes in 
> this PR calls canSeek() even in readNBytes() and skip() implementations which 
> may not be reading till EOF and in theory could be invoked multiple times.
> Having said that, the proposal to cache was merely a suggestion. I don't have 
> any data to suggest if these methods are called frequent enough to have the 
> additional native call show up prominently in that call path.

> [...] do you think we should reduce these native calls in this change and 
> call `canSeek()` just once [...]

I think that is a good idea but I would think it best to do i in a similar way 
to the recent change to `ChannelInputStream` that added `isOther`.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/21673#discussion_r1815201732

Reply via email to