Cmdstanr: cmdstan_path() set but I get a message about it not being set

[bbbales2]

Hit a weird issue this morning where my path is set but when I try to build a model it tells me it isn't:

> library(cmdstanr)
This is cmdstanr version 0.2.0
- Online documentation and vignettes at mc-stan.org/cmdstanr
- CmdStan path set to: /home/bbales2/.cmdstanr/v2.25.0-rc1
- Use set_cmdstan_path() to change the path
> mod = cmdstan_model("bee.stan")
Compiling Stan program...
Error: CmdStan path has not been set yet. See ?set_cmdstan_path.
> cmdstan_path()
[1] "/home/bbales2/.cmdstanr/v2.25.0-rc1"
> set_cmdstan_path(cmdstan_path())
CmdStan path set to: /home/bbales2/.cmdstanr/v2.25.0-rc1
> mod = cmdstan_model("bee.stan")
Compiling Stan program...
Error: CmdStan path has not been set yet. See ?set_cmdstan_path.

That folder exists on my computer and appears to have a cmdstan in it.

[rok-cesnovar]

We debugged this with Ben in a chat. The problem was in the -rc release. Guess I can close?

[bbbales2]

Well I'd prefer it keeps running and spits out a warning or something.

I suppose an error would be better than just not working with cmdstan missing.

Why did the rc work in the first place, for instance? When rc 2.26-rc1 comes along will it work?

[jgabry]

Weird, I didn't have any issue using the previous release candidate with cmdstanr. What did you figure out when debugging this?

[bbbales2]

The was the output of some more commands:

> set_cmdstan_path(cmdstan_path())
CmdStan path set to: /home/bbales2/.cmdstanr/v2.25.0-rc1
> cmdstanr:::.cmdstanr$PATH
[1] "/home/bbales2/.cmdstanr/v2.25.0-rc1"
> cmdstanr:::.cmdstanr$VERSION
character(0)
> cmdstanr:::read_cmdstan_version("/home/bbales2/.cmdstanr/v2.25.0-rc1")
character(0)

[jgabry]

Hmm the path you have is

/home/bbales2/.cmdstanr/v2.25.0-rc1

but by default cmdstanr would use

/home/bbales2/.cmdstanr/cmdstan-2.25.0-rc1

That said, your version should still work because we want to allow you to set the path to wherever you want, but maybe the difference is causing the error? I haven't dug into the code so I'm not sure, but it's a possibility.

[jgabry]

The problem was in the -rc release. Guess I can close?

I think we should keep it open because don't we want it to work with release candidates too?

[bbbales2]

/home/bbales2/.cmdstanr/v2.25.0-rc1

Good catch. I checked again and that folder does exist and contains stuff though.

[rok-cesnovar]

Yes, we should come to the bottom of this, but I honestly think was installed with a tarball that was later scratched because it either had the wrong name or something. We had a few times where we made a tarball but quickly saw it had the wrong name or some of the release replace-text procedures went wrong and we quickly replaced it.

I tried installing it locally on a Ubuntu/Windows and it works fine.

@bbbales2 can you check CMDSTAN_VERSION inside the Makefile in that folder?

[rok-cesnovar]

Actually run the following and let me know where it breaks:

  path <- "/home/bbales2/.cmdstanr/v2.25.0-rc1"
  makefile_path <- file.path(path, "makefile")
  if (!file.exists(makefile_path)) {
    warning(
      "Can't find CmdStan makefile to detect version number. ",
      "Path may not point to valid installation.",
      call. = FALSE
    )
    return(NULL)
  }
  makefile <- readLines(makefile_path)
  version_line <- grep("^CMDSTAN_VERSION :=", makefile, value = TRUE)
  sub("CMDSTAN_VERSION := ", "", version_line)

[bbbales2]

It appears to not have a CMDSTAN_VERSION:

grep("^CMDSTAN_VERSION :=", makefile, value = TRUE) returns character(0)

[rok-cesnovar]

Yeah, that was a botched RC release try then and that was updated immediately. You were an early tester :) We rely on CMDSTAN_VERSION being there. If it isn't, we cant do anything else.

[jgabry]

In that case I'm not worried about this and I think we can close it. Although maybe we should add a check for this to read_cmdstan_version(). Something like this?

  if (length(version_line) == 0) {
    stop("CmdStan makefile is missing a version number.", call. = FALSE)
  }

[rok-cesnovar]

Let's add that. Would serve as a release-checker as well.