21100 - Implement file cache

Status: VERIFIED FIXED

(Core :: Networking: Cache, defect, P3)

Description

[Scott Furman]

Placeholder bug for the file cache:

The file cache is largely done (written by [e-mail]), but it needs
final review and testing.  Once the code is checked in, the ifdef
FILE_CACHE_IS_READY lines in nsCacheManager.cpp will need to be enabled.

Comment 1

[leger]

Bulk move of all Cache (to be deleted component) bugs to new Networking: Cache
component.

Comment 2

[Warren Harris]

Assigning fur's cache bugs to Gordon. He can split them up with davidm.

Comment 3

[Yixiong Zou]

The nsIFileSpec is phasing out right? Should we change the Cache API 
accordingly? 

Comment 4

[Biswapesh Chatterjee]

Will it be in for M14 ? This is bugging a lot of people. Problems include:
1) Fetching linked documents repeatedly (CSS, JS, etc)
2) 'View Source' and 'Save Page' fetching the document again.
etc.

Comment 5

[gordon]

*** Bug 26470 has been marked as a duplicate of this bug. ***

Comment 6

[Warren Harris]

David is doing this now.

Comment 7

[Simon Fraser [no longer active]]

Diff to fix DBM file creation for Mac:

Index: hash.c
===================================================================
RCS file: /cvsroot/mozilla/dbm/src/hash.c,v
retrieving revision 3.10
diff -c -2 -r3.10 hash.c
*** hash.c	2000/01/26 01:35:12	3.10
--- hash.c	2000/02/17 04:21:38
***************
*** 52,55 ****
--- 52,56 ----
  #ifdef macintosh
  #include <unix.h>
+ #include <Errors.h>
  #else
  #include <fcntl.h>
***************
*** 168,172 ****
  
  	new_table = 0;
! 	if (!file || (flags & O_TRUNC) 	|| (stat(file, &statbuf)  && (errno == 
ENOENT))) 
  	{
  		if (errno == ENOENT)
--- 169,179 ----
  
  	new_table = 0;
! 	if (!file || (flags & O_TRUNC) 	|| (stat(file, &statbuf) &&
! #if defined (macintosh)
!   (errno == fnfErr)
! #else	
! 	(errno == ENOENT)
! #endif
! 	)) 
  	{
  		if (errno == ENOENT)

Comment 8

[Steve Morrison]

Attachment: implement nsLocalFileUnix::GetParent, used to create cache directories. Also clean up directory permissions. This gets disk cache working on linux.

Comment 9

[Brendan Eich [:brendan]]

Here are a few comments on the getParent patch, which I hope are helpful (I hope 
my Unix knowledge isn't too rusty):

! 
!     nsCString parentPath = mWorkingPath;

Should use nsCAutoString here to avoid a malloc for short-ish (<64 char) paths.

! 
!     PRInt32 offset = parentPath.RFindChar('/');
! 
!     if (offset == -1)
!         return NS_ERROR_FILE_UNRECOGNIZED_PATH;
! 
!     parentPath.Truncate(offset);

Unix allows multiple slashes between component (or did, when I used to hack on 
namei, aka lookuppn or pn_lookup).  To handle that, you might do the RFind, then 
decrement offset and check whether it is < 0 or the char at offset is not a 
slash.  My nsStr*.h memory is too poor to sketch the code, but it should be not 
too bad.

One tricky boundary: naive code would make "" be the parent path of "/", but in 
Unix (AFAICR), "" means "." (current working directory).  For the root directory 
I think getParent should return the selfsame root directory, to match the hard 
link ".." in that dir.

! 
!     nsCOMPtr<nsILocalFile> localFile;
!     nsresult rv =  NS_NewLocalFile(parentPath.GetBuffer(), 
getter_AddRefs(localFile));
!     
!     if(NS_SUCCEEDED(rv) && localFile)

This is good belt-and-braces sanity checking, but I'm curious whether 
NS_NewLocalFile is spec'd to return non-null localFile on successful rv code?

!     {
!         return localFile->QueryInterface(NS_GET_IID(nsIFile), 
(void**)aParent);
!     }
!     return rv;

Comment 10

[ekrock's old account (dead)]

Clearing PDT- to trigger re-evaluation by PDT team based on the following new 
information: 27763 (which is beta1 PDT+) now depends on 21100, because according 
to av, gagan and amusil said that we rely on disk cache to get 
the notification needed to fix 27763. (And 27763 is a blocker for Java support 
in beta1.) If this information about the dependency of 27763 on 21100 is 
correct, then 21100 should be beta1 PDT+ so long as 27763 is. (By the transitive 
rule of PDT+ for dependencies.)

Comment 11

[leger]

Spoke with warren, he will investigate to give PDT more info.

Comment 12

[leger]

Putting on PDT- radar for beta1.  No time left per phil.

Comment 13

[Warren Harris]

Marking beta2.

Comment 14

[Warren Harris]

Moving to M15.

Comment 15

[Warren Harris]

Moving what's not done for M15 to M16.

Comment 16

[davidm]

Turned on the cache marking fixed will open new bugs for any problems

Comment 17

[Tom Everingham]

marking verified - file cache is in