Problem creating FRXDO from frx: Index: 395

Problem

There is no bounds checking when mapping gutter comments to lines of code.
This means an entire review explodes with the following error, when the comments somehow don't match the lines in the file.
We should handle this more gracefully.

2011-07-13 23:57:34,151 ERROR [btpool0-6324 ] fisheye.app ViewFRXAction-execute - Problem creating FRXDO from frx
java.lang.IndexOutOfBoundsException: Index: 395, Size: 386
	at java.util.ArrayList.RangeCheck(ArrayList.java:547)
	at java.util.ArrayList.get(ArrayList.java:322)
	at com.cenqua.crucible.view.FRXDO.mapGutterComments(FRXDO.java:861)
	at com.cenqua.crucible.view.FRXDO.mapInlineComments(FRXDO.java:768)
	at com.cenqua.crucible.view.FRXDO.<init>(FRXDO.java:198)
	at com.cenqua.crucible.revision.managers.DefaultContentManager.makeFRXDO(DefaultContentManager.java:104)
	at com.atlassian.crucible.actions.ReviewBaseAction.makeFRXDO(ReviewBaseAction.java:596)
	at com.atlassian.crucible.actions.ViewFRXAction.execute(ViewFRXAction.java:184)
        ...

Most probable reason

The root cause is not in line calculation algorithm, but in an inconsistent data store.
When users move/restore backup of Crucible instance incorrectly, we end up with DB out of sync with the files. This can result in db reusing unique file identifiers and Crucible overwriting a file in its cache. Where we used to have file A, now we have file B. Yet we still think we have file A. In effect, when we load a review for file A, we show contents of file B. If there has been a comment previously added to file A, say on line 395, and file B is only 386 lines long, we throw IndexOutOfBoundsException when we try to place that comment on file B.

Workaround

This workaround will only work if the source repository is available and running. It will fail if repository is no longer available (including stopped/disabled) or the file originally came from an attachment.

On *nix you can perform this fix while the instance is running (not sure about Windows).

Find CFR number

Go to the file that doesn't load properly. CFR is displayed on the address bar after the '#'. Copy this number.

SQL query to find the file names

Run the following query, substituting <CFR> with the number you have just copied:

SELECT
    r.cru_revision AS revision,
    r.cru_upload_item AS upload_item
FROM
    cru_frx_revision fr,
    cru_revision r
WHERE
    fr.cru_revision = r.cru_revision_id
AND fr.cru_frx_id = <CFR>
ORDER BY r.cru_upload_item

Delete cache files

The last column (upload_item) gives a list of cache file ids used to display this file (one per revision). Delete files named <upload_item>.dat from these folders:

• $FISHEYE_INST/var/tmp/encodedcontent/uploaditem
• $FISHEYE_INST/var/data/uploads

Reload review page

Refresh the review twice. You should see the correct files loaded from the repository.

Workaround - example

In my case the address bar shows http://localhost:8080/cru/CR-3#CFR-7, so CFR = 7. The SQL query returns

revision upload_item 
-------- ----------- 
2000     19          
1999     20          
31       23          
30       24          
210      25          
211      26

Then I look for these files:

$ for CACHE_FOLDER in ./var/tmp/encodedcontent/uploaditem ./var/data/uploads/; do for UPLOAD_ITEM in 19 20 23 24 25 26 ; do find $CACHE_FOLDER -name $UPLOAD_ITEM.dat; done; done
./var/tmp/encodedcontent/uploaditem/00/00/19.dat
./var/tmp/encodedcontent/uploaditem/00/00/20.dat
./var/tmp/encodedcontent/uploaditem/00/00/23.dat
./var/tmp/encodedcontent/uploaditem/00/00/24.dat
./var/tmp/encodedcontent/uploaditem/00/00/25.dat
./var/tmp/encodedcontent/uploaditem/00/00/26.dat
./var/data/uploads//00/00/19.dat
./var/data/uploads//00/00/20.dat
./var/data/uploads//00/00/23.dat
./var/data/uploads//00/00/24.dat
./var/data/uploads//00/00/25.dat
./var/data/uploads//00/00/26.dat

and delete all of them. Refresh browser twice and voilà!

Proper fix

Rather than using consecutive numbers as file names, use their hash instead. When loading a file, check the hash matches. If not, invalidate the file store and pull the file from the repository.