I'm attempting to script the update of an existing asql database with new
log entries, but re-saving the database file results in an empty log file.
The problem isn't automation specific, it'll blow up quite happily
interactively to, as the following snippets can demonstrate:
----->8-----
$ cat logs1
[1000 lines of log file]
$ cat logs2
[another 1000 lines of log file]
$ rm -f logs.sql
$ asql
asql v1.6 - type 'help' for help.
asql> load logs1
Loading: logs1
asql> save logs.sql
Saving to : logs.sql
asql> quit
$ ls -l logs.sql
-rw-r--r-- 1 mpalmer mpalmer 326656 Oct  1 15:11 logs.sql
$ asql
asql v1.6 - type 'help' for help.
asql> restore logs.sql
asql> load logs2
Loading: logs2
asql> save logs.sql
Saving to : logs.sql
asql> quit
$ ls -l logs.sql
-rw-r--r-- 1 mpalmer mpalmer 0 Oct  1 15:11 logs.sql
-----8<-----

I thought that perhaps one needed to avoid giving a filename to 'save' if
you'd restored, so I tried that, too:

----->8-----
[as above, up to:]
asql> load logs2
Loading: logs2
asql> save
Saving to : /home/mpalmer/.asql.db
asql> quit
$ ls -l /home/mpalmer/.asql.db
-rw-r--r-- 1 mpalmer mpalmer 0 Oct  1 15:18 /home/mpalmer/.asql.db
$ ls -l logs.sql
-rw-r--r-- 1 mpalmer mpalmer 651264 Oct  1 15:11 logs.sql
-----8<-----

So, if I save a database that previously had some content to a different
location (either the default given by 'save' without an argument, or a
different location I explicitly specified -- I tried both), then I get an
empty database in the new location, and the database I restored from gets a
whole pile of extra records.

I can live with the latter behaviour, if it's well-documented, but I think
that a 'save' causing an empty database anywhere is pretty nasty.

My suggested fixes for this would be:

* Fix save so that re-saving an existing database (either to the existing
location or a new one) doesn't produce an empty database.

and one of either:

* Documenting the behaviour of the 'restore' command to state that restoring
  a database effectively makes it the 'live' database for all subsequent
  load commands (and that a further 'save' is unnecessary); or

* Modifying the 'restore' command to make a copy of the restored database to
  a separate location, and then having the 'load' and 'select' commands
  operate on that database instead of the file that was restored from, and
  requiring an explicit 'save' command to persist that modified database
  (this could be expensive in disk space for large DBs, I know).

If any of the current behaviours are documented, then I missed them -- I
checked asql(1), README, and the output of 'help', 'help restore', 'help
load', and 'help save'.

Thanks,
- Matt