802107 - [OS.File] Copy files with sendfile wherever splice is not available

Status: RESOLVED WONTFIX

(Toolkit Graveyard :: OS.File, enhancement)

Description

[David Teller [:Yoric] - still alive but not very active]

Under Linux/Android/Firefox OS, most file systems support operation |sendfile()|, which is generally the fastest way of copying the contents of a file. We should implement this.

Comment 1

[Gabriel Corona]

Is sendfile really faster than splice (which is implemented)? With 2.6.33 and above, sendfile uses splice internally (in the kernel).

Comment 2

[David Teller [:Yoric] - still alive but not very active]

Good point. Before making any robust implementation, it would be useful to benchmark a |sendfile|-based implementation vs. a |splice|-based implementation on various configurations.

Comment 3

[Mike Hommey [:glandium]]

(In reply to David Rajchenbach Teller [:Yoric] <away until Jan 7th> from comment #2)
> Good point. Before making any robust implementation, it would be useful to
> benchmark a |sendfile|-based implementation vs. a |splice|-based
> implementation on various configurations.

As mentioned in comment 1, sendfile uses splice internally. No amount of benchmark will show significant difference, since they are the same thing. The advantage of sendfile over splice is that it is available on FreeBSD and Solaris, iirc.

Comment 4

[David Teller [:Yoric] - still alive but not very active]

Ah, right. I replied without thinking. Updating the topic.

Comment 5

[David Teller [:Yoric] - still alive but not very active]

Any news?

Comment 6

[David Teller [:Yoric] - still alive but not very active]

On second thought, the patch would have an extremely minor impact and would not even be tested by Try, so this is probably not the right place to spend our time and efforts.

Comment 7

[Mike Hommey [:glandium]]

(In reply to David Rajchenbach Teller [:Yoric] from comment #6)
> On second thought, the patch would have an extremely minor impact and would
> not even be tested by Try

Why wouldn't it?

> so this is probably not the right place to spend our time and efforts.

If a volunteer wants to come in and do it, i don't see why we should prevent him from doing so. Mitesh is a volunteer, isn't he?

Comment 8

[David Teller [:Yoric] - still alive but not very active]

(In reply to Mike Hommey [:glandium] from comment #7)
> (In reply to David Rajchenbach Teller [:Yoric] from comment #6)
> > On second thought, the patch would have an extremely minor impact and would
> > not even be tested by Try
> 
> Why wouldn't it?

Last time I checked, we didn't have FreeBSD or Solaris on Try.

> > so this is probably not the right place to spend our time and efforts.
> 
> If a volunteer wants to come in and do it, i don't see why we should prevent
> him from doing so. Mitesh is a volunteer, isn't he?

Sure, but as the mentor for that bug, I recommended that he should find a more interesting mentored bug to work on :)

Comment 9

[Marco Castelluccio [:marco]]

Isn't the bug still valid? I mean, if someone comes with a patch we wouldn't reject it.

Comment 10

[David Teller [:Yoric] - still alive but not very active]

Fair enough, reopening but de-mentoring.

Comment 11

[Beth Rennie [:beth] (she/her)]

OSFIle is being replaced with IOUtils and PathUtils.