Parallel ado path error

[Felix Bittmann]

Dear all,
I would like to use the ado parallel on a virtual machine (remote data access machine due to security issues of the data provider). The machine runs Windows 7, Stata 16.1 and the ado is installed (1.20.1). However, as soon as I enter

parallel initialize 4

I get the following error message:

parallel_setstatapath(): 3499 default_statapath() not found
<istmt>: - function returned error

So I tried to set the path like

parallel initialize 4, statapath("C:\Program Files\Stata16\StataMP-64.exe")

I get the same error. I checked, the path is correct and the .exe is there (using pwd and cd).

Any ideas what is wrong here?

[daniel klein]

The problem is not with the path or exe-file; the problem is that Mata cannot find the function default_statapath().

How did you install parallel? The package appears to contain (much) more than one ado-file, and the version that you report (1.20.1) is not available from SSC. My guess is that you have installed only parts of the parallel package and/or rely on an outdated (or buggy) Mata library file. It is hard to say exactly what is going on. Try re-installing from SSC and/or contact the author of parallel.

[Daniel Bela]

Hi Felix Bittmann,

this happened to me (and our IT department – the "data provider" you mentioned ) recently as well. The problem, in our case, was that it seems the latest version in the master branch of parallel's GitHub repository seems to have a bug with regards to detecting Stata's default path.

We solved the problem by using the stable branch as installation source, like in:

net install parallel, from(https://raw.github.com/gvegayon/parallel/stable/)
mata mata mlib index

Note that the authors explicitly recommend not to use the older version from SSC, unless you need to, as it doesn't contain recent fixes.

All the best
Bela

[Daniel Bela]

I also informed the authors about the issue on GitHub now. The best workaround, imho, is using stable instead of master.

Kind regards
Bela

[Brian Quistorff]

Thanks for reporting this issue to the GitHub site (sorry, I don't monitor the forum very much). There was an error with copying one function into the mlib file. It should be fixed now in master.