119350 - MsgNavigation does not follow top-down order in some cases.

Status: VERIFIED FIXED

(SeaMonkey :: MailNews: Message Display, defect, P3)

Description

[Navin Gupta]

I have seen some issues here, will file new bug about this. looks
like we might need to do CompareRawSortKey when comparing folder keys 
rather than relying on js (key1 <key2)

fix in hand, I will attach a patch.

Comment 1

[Navin Gupta]

Attachment: proposed fix

the fix as i said is to use CompareRawSortKey rather than key1 <key2. 

seth, please review.

Comment 2

[(not reading, please use [e-mail] instead)]

didn't we switch over to ascii safe collation keys?

Did we switch back because of the macos x problems, or did we never switch, and 
I'm just confused?

If we did switch to ascii safe keys, the js check would have worked.  

but since we're doing raw sort keys, we have to use the raw sort key compare.

this patch basically treats the sort keys are PRUnichar *s, and then you cast 
to PRUint8 *.

I thought that was bad, or I remember some comments in the code (from nhotta?) 
about why it was bad.

nhotta, can you clarify?

Comment 3

[Navin Gupta]

It was bad to use PRUnichar*, but we were having problems because rdf
wanted them to be PRUnichar*, so we decided to go this way. I believe
there are lots of bug filed to make all this work correctly. 

This is how OutlinerBuilder does sorting in the folder pane ie casts
from PRUnichar* to PRUint8*

Comment 4

[nhottanscp]

I did not check in the ASCII encoded key support. Providing more than one key
format could cause confusion because the client code has to use different
comparison functions and they are easily be mixed up and causes wrong results. 
Also the current implementation in XUL is fixed to one comparison code and
cannot be changed by the caller, so if one client of the XUL sort switch to
ASCII key then all the ther clients has to switch to ASCII keys. And ASCII keys
solution needs encode/decode, encode might be not significant only once per key,
but decode has to be done per comparison. The current plan is to change XUL sort
to support binary key which is bug 115926.

Comment 5

[[e-mail]]

Comment on attachment 64401 [details] [diff] [review]
proposed fix

If you're abandoning .sortKey you could remove that from the .idl and
associated files.

Comment 6

[Navin Gupta]

we still need that because we are asking for sortKey from msgFolderDataSource.
i think we should let it be scriptable. 

Comment 7

[(not reading, please use [e-mail] instead)]

I agree with neil, don't make it scriptable.

Casting the raw sort keys from  PRUnichar * to PRUint8 * is evil, as we know.

let's not expose that through XPConnect.

When crossing XPConnect, we copy the strings, and probably call strlen() during 
that process.  That means null bytes in the raw sort key will mess us up.

Let's not expose this evil to JS if we don't have too.

can you amend the patch to make it nonscript?

nhotta, can you provide the r=?

Comment 8

[nhottanscp]

So the patch has a comparison code in the mail code, does this mean this part
does not use RDF? Is this for a folder tree sorting?

Comment 9

[Navin Gupta]

This is just for msgNavigation so that it follows the order in which
the folders appear in the folder-pane. 

Comment 10

[nhottanscp]

So msgNavigation is that the list I can see when I move/copy message?

Comment 11

[Navin Gupta]

Attachment: proposed fix

same patch with sortKey non-scriptable.

Comment 12

[Navin Gupta]

msgNavigation is "Next" in the toolbar. so when you click next it 
should follow top-down order in the folder-pane, as it navigates
into folders with unread messages. 

Comment 13

[nhottanscp]

Comment on attachment 64574 [details] [diff] [review]
proposed fix

r=nhotta

Comment 14

[(not reading, please use [e-mail] instead)]

Comment on attachment 64574 [details] [diff] [review]
proposed fix

looks good.

one suggestion (and brendan would be so proud of me)

+ nsresult rv=NS_OK;

no need to initialize this value.

Comment 15

[Navin Gupta]

fixed

Comment 16

[Navin Gupta]

To verify this create two local folders '34' and 'aaa' and have unread messages 
in both and see the difference before and after the fix. 

Comment 17

[laurel]

OK using feb04 commercial trunk build: win98, mac OS 10.1, linux rh6.2