read csv with variable names that contain parentheses

[yizhang-yiz]

Describe the bug I'm trying to piggyback cmdstanr to read samples generated elsewhere. as_cmdstan_fit complains variable names such as "OMEGA(1,1)", whiledata.table::fread ingests fine. I think it's caused by the ~~internal fread_cmd's grepping~~ unpacking variable names.

To Reproduce For a csv with content (dummy metadata)

# model = norm_model
# start_datetime = 2025-11-19 22:23:43 UTC
# method = sample (Default)
#   sample
#   ....
THETA4,"SIGMA(1,1)"
2.00000E+00,2.00000E+00
2.00000E+00,2.00000E+00

as_cmdstan_fit("foo.csv", check_diagnostics=FALSE) errors

Error in grep(pattern, var_names) : 
  invalid regular expression '^"SIGMA(1\[', reason 'Missing ')''
In addition: Warning message:
In grep(pattern, var_names) : TRE pattern compilation error 'Missing ')''

Expected behavior No error, as by fread:

fread(cmd="grep -v '^#' --color=never foo.csv")
   THETA4 SIGMA(1,1)
    <num>      <num>
1:      2          2
2:      2          2

Operating system MacOS 15.7 (24G222) 11881.140.96 Ubuntu 24.04

CmdStanR version number cmdstanr_0.9.0 & master 7c7b5d2b

[jgabry]

@yizhang-yiz Thanks for reporting this. Does #1120 solve this for you?

[jgabry]

The issue is a bit more complicated than I thought. We can get fread working using #1120, but unfortunately there are some other problems after that that I don't have time to dive into (I was hoping #1120 would be sufficient). I think for now the best solution is to convert variable names before calling as_cmdstan_fit(). If you have a variable name Sigma(1,1) then CmdStan would name it Sigma.1.1. If you replace the parentheses with periods using regular expressions or whatever method you want then it should work with as_cmdstan_fit(). In the resulting fitted model object it would then become Sigma[1,1].

EDIT: maybe we can make this work after all. See discussion over at #1120