Error and crashes of autotest::autotest_package

[KlausVigo]

Hello,

I try to run autotest::autotest_package("phangorn") and either this kills the RStudio session or I get different error messages.
The number of examples and .Rd files it tries to extract information seem also to differ, e.g. adding or deleting S3 methods on the fly.
I only can pinpoint it down to Rd file which document several functions or contain S3 methods. roxygen2 seems to change these files also for whatever reason frequently.

> autotest::autotest_package("phangorn")
★ Extracting example code from 66 .Rd files
  |===========================================================================| 100%
✔ Extracted example code
★ Converting 26 examples to yaml
  |===========================================================================| 100%
✔ Converted examples to yaml

── autotesting phangorn ──

✔ [1 / 30]: add.tips
✔ [2 / 30]: as.splits
✔ [3 / 30]: Ancestors
✔ [4 / 30]: allTrees
✔ [5 / 30]: as.ancestral
✔ [6 / 30]: as.networx
✔ [7 / 30]: cladePar
✔ [8 / 30]: coalSpeciesTree
✔ [9 / 30]: dfactorial
✔ [10 / 30]: discrete.gamma
✔ [11 / 30]: dist.p
✔ [12 / 30]: getClans
✔ [13 / 30]: getDiversity
Fehler in strsplit(yaml2[grep(ystr, yaml2)], ystr)[[1]] : 
  Indizierung außerhalb der Grenzen

or without "as.ancestral"

> at <- autotest::autotest_package("phangorn")
★ Extracting example code from 66 .Rd files
  |===========================================================================| 100%
✔ Extracted example code
★ Converting 25 examples to yaml
  |===========================================================================| 100%
✔ Converted examples to yaml

── autotesting phangorn ──

✔ [1 / 29]: add.tips
✔ [2 / 29]: as.splits
✔ [3 / 29]: Ancestors
✔ [4 / 29]: allTrees
✔ [5 / 29]: as.networx
✔ [6 / 29]: cladePar
✔ [7 / 29]: coalSpeciesTree
✔ [8 / 29]: dfactorial
✔ [9 / 29]: discrete.gamma
✔ [10 / 29]: dist.p
✔ [11 / 29]: getClans
✔ [12 / 29]: getDiversity
Fehler in strsplit(yaml2[grep(ystr, yaml2)], ystr)[[1]] : 
  Indizierung außerhalb der Grenzen

Ad here is the output of traceback():

> traceback()
11: is.factor(x)
10: gsub("^\\s+", "", strsplit(yaml2[grep(ystr, yaml2)], ystr)[[1]][2])
9: parse_one_fn(x, f, yaml)
8: parse_yaml_template(yaml = yaml, filename = filename)
7: autotest_single_yaml(i, filename, test, test_data, quiet)
6: FUN(X[[i]], ...)
5: lapply(yaml, function(i) autotest_single_yaml(i, filename, test, 
       test_data, quiet))
4: autotest_yaml(yaml = yaml, test = test, test_data = test_data, 
       quiet = TRUE)
3: rbind(deparse.level, ...)
2: rbind(res, autotest_yaml(yaml = yaml, test = test, test_data = test_data, 
       quiet = TRUE))
1: autotest::autotest_package("phangorn")

Kind regards,
Klaus

[mpadge]

Hi @KlausVigo, and thanks as always for asking. This time I've only got bad news for you, I'm afraid. I can also reproduce what you're seeing, but it is extremely unpredictable, and rarely reproduces in the same way. I roughly understand where the problem is arising. It's definitely in this huge file, and somewhere in one of these steps here. The problem is that I can't reproduce it in any way reliable enough to understand what is happening. I've stripped out all code beyond that point, and got a version that will run one time, but then error on a second identical call. Even then, it sometimes just aborts and kills the R process, and other times the process remains alive and generates an error trace. And then even when that happens, the traces are not reproducible...

The larger problem is that we ran out of funding for further development of this package a while ago. The whole package has been left hanging with partial solutions to most problems still in #78. That will hugely simplify most aspects, and entirely remove all those manual parsing routines that are causing problems like these. I'll have a think about whether we might be able to prioritize this package a bit, and get things moving again. In the meantime, I suggest the only option for your {phangorn} package is to ignore autotest for the moment, unfortunately. Or, if you really want to autotest it, try coping two versions, each with only half of the .Rd files in the /man directory. That also seems to work well. Please accept my sincere apologies for this inconvenience, and let me know if there's anything more I can do to help, or to investigate further.