This is a known limitation of valiases. I've wanted to do something about it for a long time, but don't have the spare time to do so. It would be great if John and perhaps others could pitch in to make it happen.

On Jul 19, 2005, at 3:57 PM, John Simpson wrote:
i don't know that i'll have the free time, but i'm certainly willing to look at it.


so what's actually needed would be:

- a sequence field in the table, integral type, 32-bit unsigned integer (i would say limit it to 8 or 16 bits, but you never know...)

- vdelivermail.c needs an ORDER BY clause. it may also need to be modified to treat exit code 99 from a program delivery the same way qmail-local does (i.e. treat it as a success but stop processing alias lines) unless i'm missing something and it already does this?

Actually, vdelivermail won't require changes. Just update valias_select() in vmysql.c

- need a script to convert a non-sequence table into a sequence table. this involves creating a new table with the sequence field, copying the old table to the new one (manually creating the sequence for all entries), deleting the old table, and renaming the new one to the old name. it could also identify any "problem" mailboxes- those with multiple lines, at least one of which is a program delivery.

You could just add the column to the existing table and then populate it with 0. I agree that a tool to identify problematic aliases would be good, but setting it to 0 would match current behavior.

- valias.c needs the same ORDER BY clause as vdelivermail.c (for use when listing aliases.) we may also want a "-n" command line flag which makes "-s" show the seqence numbers in addition to the current output (but don't change the output, that would break scripts which depend on this output.)

Sounds good. You might need to modify linklist_add() in vpopmail.c and update the calls to it in vmysql.c to store the sequence number.

- the valias_insert() functions in vmysql.c, vpalias.c, and vpgsql.c need a fourth parameter, a sequence number (unsigned). technically it's how many lines to skip before adding the new entry- 0 means "add at the beginning", 1 means "skip the first line and then insert the new one", and so forth... anything >= number of lines in the file is treated as "add to the end". the functions need to be able to re-number the existing rules in the database as needed to maintain the correct sequence.

Like this?

update valias set seq = seq + 1 where alias = '[EMAIL PROTECTED]' and seq > 4 insert into valias (alias, alias_line, seq) values ('[EMAIL PROTECTED]', '[EMAIL PROTECTED]', 4)

- dotqmail2valias.c needs to make all of its calls with -1 as the sequence, meaning that all rules added would be added to the end of whatever instructions are already there for that alias.


- valias.c would use the new "-n" command line option to specify the sequence for the new rule added by the "-i" option. (i'm assuming that "-i" and "-s" are mutually exclusive in the code.)

Sounds good to me.

am i missing anything?

Very thorough. We'd have to update qmailadmin as well, but since it presently always adds to the end, it wouldn't be a big change. Ultimately, we would want to update the UI in QmailAdmin to allow you to change the order of lines in an alias (kind of like how Netflix manages your queue).

Perhaps when creating the new valias_insert() command, we should give it a new name and keep valias_insert() with the same parameters. valias_insert() would just call the new command and pass in -1 for the sequence. That way, old programs wouldn't break.

We might also want to define a macro to indicate that the extended insert is available, so programs like QmailAdmin can have conditional compilation depending on whether they're linked to the old or new vpopmail.

Tom Collins  -  [EMAIL PROTECTED]
QmailAdmin:  Vpopmail:
You don't need a laptop to troubleshoot high-speed Internet:

Reply via email to