167525 - When 1024 byte page is in cache, Error message "The file /test/ cannot be found. Please check the location and try again"

Status: VERIFIED FIXED

(Core :: Networking: Cache, defect)

Description

[James Paige]

Linux trunk build 2002090905

Load the url http://www.westcoastaerospace.com/test/

It will load just fine

Now hit the reload button, or press back and then press forward, or hit enter in
the URL bar, or perform any other action that causes the page to be reloaded.

You will see the error message "The file /test/ cannot be found. Please check
the location and try again"

If you press Shift+Reload to bypass the cache, the page will reload just fine.
If you go into preferences and clear your disk and memory caches, you will be
able to reload the page again once.

For similar symptoms see also:
 bug 141831 (same symptom, but with ftp)
 bug 123808 (marked worksforme after resolution of bug 149090)
 bug 155973 (same symptom as a result of deleting your cache dir)

[Also, note that the links on http://www.westcoastaerospace.com/test/ are
supposed to be broken. That index.html file is just one single file from a
larger body of documentation on an internal server that I cannot make
world-accessable.]

Comment 1

[Jo Hermans]

Confirmed on build 2002090408 on Mac OS 9
so Platform/OS --> All

I can't see anything special with the HTTP-sniffer at
http://webtools.mozilla.org/web-sniffer/ . I tried the same URL in Internet
Explorer 5 and iCab, but it works ok.

Comment 2

[Jean-Pierre Melkonian]

view source has the same problem
Changing character coding also
(Mac OS 9.1, Build 2002090608)
Works with NS4

Comment 3

[Luke Koleszar]

Tried with some older trunk versions (All WinME)
2002090904, 2002090404, 2002090208, 2002082804 all have the problem

1.1/20020826 works as expected

Comment 4

[James Paige]

Aha! I think I have found something important. I was frustrated to find that
adding or removing even a single letter of that index.html page caused the
problem to go away (I was trying to correct the spelling of "Rebulding") and
then I realized that this particular file is *exactly* 1024 bytes.

Comment 5

[James Paige]

works   http://www.westcoastaerospace.com/test/1023.txt   1023 bytes
fails   http://www.westcoastaerospace.com/test/1024.txt   1024 bytes
works   http://www.westcoastaerospace.com/test/1025.txt   1025 bytes

Comment 6

[James Paige]

512 byte file http://www.westcoastaerospace.com/test/512.txt does not exhibit
the problem.

Comment 7

[James Paige]

neither does 2048 byte file http://www.westcoastaerospace.com/test/2048.txt

Only 1024 byte files seem to have this problem. Amending summary.

Comment 8

[Luke Koleszar]

Attachment: 4096 byte file also having the cache problem

Comment 9

[Luke Koleszar]

Attachment: 16K file that also has the cache problem

Comment 10

[Luke Koleszar]

After seeing what seems like the same problem on a couple of real-life web pages
(http://slashdot.org/index.pl, http://c2.com/cgi/quickChanges?days=2) I did a
little more testing of various size files.  I found that 4096 and 16384 byte
files also have the same problem.  The next value in sequence 65536 works fine
as well as 8K and 32K.

To reproduce the problem:
1) Click on attachment
2 [review]) Click on bookmark or enter new URL
3) Hit back button, see error message

(Tested with build 2002091608 WinME)

Comment 11

[gordon]

1k, 4k, and 16k are decision points in where to store cache entry data.  It's
not entirely surprising we break there.  I bet we got the math comparisons
wrong.  If I'm right, it shouldn't be too hard to fix.  Nice catch.  Thanks.

Comment 12

[gordon]

Attachment: fix 1k, 4k, 16k boundary problem

The problem occurred when trying to write the cache blocks. The way the cache
block count was calculated had an off by one bug in it, in relation to
CalculateFileIndex().

Comment 13

[Doug Turner (:dougt)]

Comment on attachment 102205 [details] [diff] [review]
fix 1k, 4k, 16k boundary problem

r=dougt.  looks fine.

Comment 14

[Darin Fisher]

Comment on attachment 102205 [details] [diff] [review]
fix 1k, 4k, 16k boundary problem

sr=darin

Comment 15

[gordon]

Fix checked in.

Comment 16

[James Paige]

verified fixed with linux trunk build 2002100911