allow multiple .dyna files on command line

[jeisner]

jasoneis@a14:~$ ls *5.dyna
anagram5.dyna  subst5.dyna  swap5.dyna  swapadj5.dyna
jasoneis@a14:~$ dyna -i *5.dyna
usage: interpreter.py [-h] [-i] [--plan] [-o OUTPUT]
                      [--post-process [POST_PROCESS [POST_PROCESS ...]]]
                      [--load [LOAD [LOAD ...]]]
                      [source]
interpreter.py: error: unrecognized arguments: subst5.dyna swap5.dyna swapadj5.dyna
Saving crash as jasoneis:2013-07-03:17:38:09:32508.log ...

[timvieira]

With the same interpretation as concatenating them together?

[jeisner]

Yeah, concatenated in the order given. Seems a pretty standard Unix interpretation of giving a list of files to something that reads stdin.

(There are times when one might want to load multiple dynabases under different names, but that should probably require a little extra command-line syntax to specify the names. Or a command-line flag to indicate that default dynabase names should be derived from the filenames; these names would have to be single-quoted if they are not ok as identifiers. We could also define aliases of the form $loaded(0), $loaded(1), ... as well as $loaded("foo.dyna"), $loaded("bar.dyna"), etc. I'm thinking that we'd like to support generic processing of a list of files; have to work out the details.)

[nwf]

Working on making the compiler understand. Amusingly enough, the parser framework does not seem to have this right; investigating.

[timvieira]

I don't understand why we want this. (Python doesn't take multiple files. gcc/javac do, but they do something different with it.)

In your example of loading multiple dynabases, you should either type those in at the repl our write a file which imports them and give them names.

[jeisner]

We don't have multiple dynabases yet (that comment was looking ahead). All we have is concatenation of stuff specified in different .dyna files. So in fact, this is a lot like providing multiple files to javac, with different parts of the program defined in different files but intended to work together.

At the moment, I want to run forward-backward (encapsulated in forward_backward.dyna) on various data and initial HMM parameters (encapsulated elsewhere). E.g.,

dyna icecream_data.dyna icecream_hmm.dyna forward_backward.dyna

Each file is short enough to be comprehensible by itself, and I can easily invoke the command line with different choices for the first two files -- trying different initial parameters, or switching from weather reconstruction to POS tagging.

Right now, I have to cat the files together and make up a name for the result and then run dyna on that, which is a pain.

Note: The load command doesn't yet make it possible to suck in new rules from a .dyna file (even though I think you can use a load command at any time to suck in new facts from any non-.dyna file). How about an "include" command for this, by analogy with C's #include? I am suggesting that .dyna files on the command line are included.

[jeisner]

Right now, I have to cat the files together and make up a name for the result and then run dyna on that, which is a pain.

Especially since error messages give line numbers in the concatenated file. They should refer to one of the original source files (which of course is impossible in the current design).