Cmdstanr: Cmdstanr not forwarding compile errors

[bbbales2]

I'm not sure if this is me or not, but the develop cmdstanr isn't forwarding me my error messages.

If I compile a model with a syntax error:

parameters {
  row_vector[3] a;
  vector[3] b;
}

model {
  matrix[1, 1] c = a * b;
}

With this R code:

library(cmdstanr)

model = cmdstan_model("/home/bbales2/tmp/test.stan")

I'm getting weird errors (instead of the syntax errors I expect):

> model = cmdstan_model("/home/bbales2/tmp/test.stan")
Compiling Stan program...
Error in processx::run(command = make_cmd(), args = c(tmp_exe, cpp_options_to_compile_flags(cpp_options),  : 
  System command error

[rok-cesnovar]

I get this (Ubuntu):

Compiling Stan program...
Error in processx::run(command = make_cmd(), args = c(tmp_exe, cpp_options_to_compile_flags(cpp_options),  : 
  System command 'make' failed, exit status: 2, stderr (last 10 lines):
E>      5:  
E>      6:  model {
E>      7:    matrix[1, 1] c = a * b;
E>            ^
E>      8:  }
E>    -------------------------------------------------
E> 
E> Ill-typed arguments supplied to assignment operator =: lhs has type matrix and rhs has type real

[bbbales2]

That's what I'd expect. Ubuntu here too. Any changes here lately? I'm pretty sure I got these errors before.

[rok-cesnovar]

Huh, no. Except for the _threads suffixes, not really. rebuild_cmdstan() might get you out of this bug, but might be worth exploring what is happening.

[rok-cesnovar]

I however don't like this format of messages. Mentioning make, processx and whatnot is probably confusing for the users. Gonna fix this real quick.

[bbbales2]

Error in processx::run(command = make_cmd(), args = c(tmp_exe, cpp_options_to_compile_flags(cpp_options), :

I agree I don't think the processx command helps. The make thing is at least giving a clue that it's compile.

[rok-cesnovar]

This is much better imo:

Compiling Stan program...
/
Semantic error in '/tmp/RtmpSiPdjL/model-692018715e71.stan', line 7, column 2 to column 25:
   -------------------------------------------------
     5:  
     6:  model {
     7:    matrix[1, 1] c = a * b;
           ^
     8:  }
   -------------------------------------------------

Ill-typed arguments supplied to assignment operator =: lhs has type matrix and rhs has type real

make: *** No rule to make target '/tmp/RtmpSiPdjL/model-692018715e71.hpp', needed by '/tmp/RtmpSiPdjL/model-692018715e71'.  Stop.
Error in self$compile(...) : 
  An error occured during compilation! See the message above for more information.

Without the E> and the rest.

I dont like the "Semantic error in '/tmp/RtmpSiPdjL/model-692018715e71.stan'" and have an idea on how to fix this. I think rstan also compiles in tmp folders and I find it slightly annoying. Opening a PR in a sec to fix the message.

[bbbales2]

Yeah you're right. No need for make

[bbbales2]

Well sometimes I go grab temporary outputs -- when I'm trying to figure out c++ toolchain problems or whatever.

[jgabry]

For what it's worth, on my mac I get this:

Compiling Stan program...
Error in processx::run(command = make_cmd(), args = c(tmp_exe, cpp_options_to_compile_flags(cpp_options),  : 
  System command 'make' failed, exit status: 2, stderr (last 10 lines):
E>      6:  
E>      7:  model {
E>      8:    matrix[1, 1] c = a * b;
E>            ^
E>      9:  }
E>    -------------------------------------------------
E> 
E> Ill-typed arguments supplied to assignment operator =: lhs has type matrix and rhs has type real
E> 
E> make: *** No rule to make target `/var/folders/h6/14xy_35x4wd2tz542dn0qhtc0000gn/T/RtmpzY8D5F/model-11a99479cfaaf.d', needed by `/var/folders/h6/14xy_35x4wd2tz542dn0qhtc0000gn/T/RtmpzY8D5F/model-11a99479cfaaf'.  Stop.

which seems to be correct. But I agree the message would look better without the processx info at the top, without the E>, and without the make stuff.

[rok-cesnovar]

Well sometimes I go grab temporary outputs

Yes, I get that. But there should be a way to override this:

Semantic error in /tmp/RtmpSiPdjL/model-692018715e71.stan, line 7, column 2 to column 25:
   -------------------------------------------------
     5:  
     6:  model {
     7:    matrix[1, 1] c = a * b;
           ^
     8:  }
   -------------------------------------------------

Ill-typed arguments supplied to assignment operator =: lhs has type matrix and rhs has type real

The first line should be "Semantic error in ~/stan-examples/eight-schools.stan" or whatever. The actual model name & path. But if I am the only one annoyed by this, its not worth it. Would be a stanc3 fix.

[jgabry]

But if I am the only one annoyed by this, its not worth it. Would be a stanc3 fix.

I find it annoying too, so would be happy to see it fixed. But not a super high priority if it's a pain to change it.

[bbbales2]

The first line should be "Semantic error in ~/stan-examples/eight-schools.stan" or whatever

Right again, that is better.

[bbbales2]

My rstudio must have been sick. I just ran this stuff again today and I got the error :/.

[rok-cesnovar]

Still got us to make the error nicer so its a win.