857830 - Consider reading ahead the spellchecker dictionary

Status: RESOLVED FIXED

(Core :: Spelling checker, defect)

Description

[(no longer active)]

We read the dictionary once but by terrible fread calls.  We should consider ReadAhead-ing it beforehand.  Aaron, are you interested in taking this?

Comment 1

[(No longer employed by Mozilla) Aaron Klotz]

Sure, I'll look into it.

Comment 2

[(no longer active)]

Thanks, let me know if you need help.

Comment 3

[(No longer employed by Mozilla) Aaron Klotz]

Attachment: Part 1: Ensure that aOutFd is always set to something when it it provided

ReadAheadFile should set aOutFd to error if the function is failing.

Comment 4

[(No longer employed by Mozilla) Aaron Klotz]

Attachment: Part 2: Readahead for the spellchecker dictionary

The entire dictionary file is being read sequentially via an fgets loop. In this patch I use the aOutFd parameter when calling ReadAheadFile, allowing me to pass the file descriptor used for readahead into fdopen, reusing it for the fgets loop.

Comment 5

[(no longer active)]

Comment on attachment 743715 [details] [diff] [review]
Part 2: Readahead for the spellchecker dictionary

Review of attachment 743715 [details] [diff] [review]:
-----------------------------------------------------------------

Please list this bug number in the "Additional patches" in README.hunspell too.

Comment 6

[Mike Hommey [:glandium]]

It would be better to avoid patching hunspell altogether if we can avoid it.

Comment 7

[(no longer active)]

(In reply to comment #6)
> It would be better to avoid patching hunspell altogether if we can avoid it.

That ship has sailed a long time ago!  This is fine since it's very well localized.

Comment 8

[Mike Hommey [:glandium]]

I'd rather see something like extensions/spellcheck/hunspell/src/hunspell_alloc_hooks.h

Comment 9

[(no longer active)]

(In reply to comment #8)
> I'd rather see something like
> extensions/spellcheck/hunspell/src/hunspell_alloc_hooks.h

Yeah you're right, that's a good idea!

Comment 10

[(no longer active)]

Comment on attachment 743715 [details] [diff] [review]
Part 2: Readahead for the spellchecker dictionary

Sorry for changing my mind!

Comment 11

[(No longer employed by Mozilla) Aaron Klotz]

Attachment: Part 2: Readahead for the spellchecker dictionary using fopen hook

This version hooks fopen via macro. It uses the readahead code path if mode is equivalent to O_RDONLY, otherwise it falls back to the C runtime's fopen.

Comment 12

[(no longer active)]

Comment on attachment 745933 [details] [diff] [review]
Part 2: Readahead for the spellchecker dictionary using fopen hook

Review of attachment 745933 [details] [diff] [review]:
-----------------------------------------------------------------

::: mozilla-config.h.in
@@ +20,5 @@
>   * unless --enable-system-hunspell is defined.
>   */
>  #if defined(HUNSPELL_STATIC)
>  #include "hunspell_alloc_hooks.h"
> +#include "hunspell_fopen_hook.h"

Nit: please name this hunspell_fopen_hooks.h for consistency.

Comment 13

[(No longer employed by Mozilla) Aaron Klotz]

Attachment: Part 2: Readahead for the spellchecker dictionary using fopen hook

Fixed nit, carrying forward r+

Comment 14

[(No longer employed by Mozilla) Aaron Klotz]

Part 1:
https://hg.mozilla.org/integration/mozilla-inbound/rev/e1f30dbb8108

Part 2:
https://hg.mozilla.org/integration/mozilla-inbound/rev/abaf789d8030

Try builds:
All other platforms:
https://tbpl.mozilla.org/?tree=Try&rev=b0f03bcc2984

Fixed for Windows:
https://tbpl.mozilla.org/?tree=Try&rev=892f1edadc45

Comment 15

[Ryan VanderMeulen [:RyanVM]]

CCing the Hunspell guys in case there's any interest in upstreaming this.

Comment 16

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/e1f30dbb8108
https://hg.mozilla.org/mozilla-central/rev/abaf789d8030