src_info() need to use the include_paths

[segasai]

Summary:

src_info() method returns incorrect results when the model has includes.

Description:

While debugging the #506 issue, I was trying to figure out what part of the compilation fails (as that's the condition when before #507 cwd was corrupted). That lead me to the src_info()
which apparently fails I think when the model has includes, because the src_info() call does not pass the include_paths to the stanc

cmdstanpy/cmdstanpy/model.py

Line 283 in 683a135

cmd = [

Here is my python program it just basically compiles a stan file (the model doesn't matter, it just needs an include of the file in the same directory)

from cmdstanpy import CmdStanModel
import os
import logging
logging.basicConfig(level=logging.DEBUG)
M = CmdStanModel('pmfit', 'models/pmfit.stan')
M.compile()

And here is the output

DEBUG:cmdstanpy:cmd: ./bin/stanc --info /home/skoposov/science/orphan2/models/pmfit.stan
cwd: /home/skoposov/soft/cmdstan-2.28.2
DEBUG:cmdstanpy:Command ['./bin/stanc', '--info', '/home/skoposov/science/orphan2/models/pmfit.stan']
	error during processing Operation not permitted
INFO:cmdstanpy:found newer exe file, not recompiling
INFO:cmdstanpy:found newer exe file, not recompiling
INFO:cmdstanpy:deleting tmpfiles dir: /tmp/tmpr54y238j
INFO:cmdstanpy:done

If I try to run --info without include-paths
I get an error

$ $HOME/soft/cmdstan-2.28.2/bin/stanc  --info /home/skoposov/science/orphan2/models/pmfit.stan
Syntax error in '/home/skoposov/science/orphan2/models/pmfit.stan', line 3, column 0, include error:
   -------------------------------------------------
     1:  functions
     2:  {
     3:  #include spline.stan
         ^
     4:  }
     5:  data
   -------------------------------------------------

While it works if include-paths are specified

$ $HOME/soft/cmdstan-2.28.2/bin/stanc --include-paths /home/skoposov/science/orphan2/models/  --info /home/skoposov/science/orphan2/models/pmfit.stan
{ "inputs": { "n_stars": { "type": "int", "dimensions": 0},
              "n_pm_nodes": { "type": "int", "dimensions": 0},
....

So my understanding is that the includes need to be passed to the src_info(). s

Additional Information:

CMDstan is provided through the environment variable and I use 2.28.2 version.
It's linux and python 3.8

Current Version:

the develop version of cmdstanpy

[WardBrian]

Thanks for the report. This is a tricky one, since we won’t necessarily know the include paths. We could make it a part of the model constructor and keep it around (instead of/in addition to the option to put it in stancflags) which also would improve the compile() experience as you wouldn’t need to supply them each time if you gave them to the model itself.

I think at the very least we can do some special casing in the function to print a better error

[segasai]

Thanks,

I think from my very cursory understanding at least this src_info() call

cmdstanpy/cmdstanpy/model.py

Line 170 in 683a135

model_info = self.src_info()

happens just after this

cmdstanpy/cmdstanpy/model.py

Line 153 in 683a135

if '#include' in program:

where the include_paths are added, so I think that information can be passed on.

[WardBrian]

Ah I had forgotten we were automatically adding the current directory as an include path. We can certainly do this in src_info as well, which should solve most of this

[segasai]

Thank you @WardBrian !