Waldo install fails with dev glue

[gaborcsardi]

I don't know where the error lies, I am merely recording this here:

* installing *source* package ‘waldo’ ...
** package ‘waldo’ successfully unpacked and MD5 sums checked
** using staged installation
** R
** byte-compile and prepare package for lazy loading
Error in get(genname, envir = envir) : object 'compare_proxy' not found
Calls: <Anonymous> ... register_S3_method_delayed -> registerS3method -> get
Execution halted
ERROR: lazy loading failed for package ‘waldo’
* removing ‘/Users/gaborcsardi/Library/R/arm64/4.4/library/waldo’
* restoring previous ‘/Users/gaborcsardi/Library/R/arm64/4.4/library/waldo’

[jennybc]

I suspect this is related to #310. @hadley do you have any thoughts?

[hadley]

I suspect you're right. Maybe this is a bug in R-devel?

[gaborcsardi]

IDK if it is a bug in base R, but it seems that this has never actually worked. We don't see errors on the CI because we don't build waldo from source, except on R-devel. But it does fail the same way in all R versions I tried, 4.4.1, 4.3.x, 4.2.x, etc. 4.0.5, and also 3.6.3:

docker run -ti ghcr.io/r-hub/r-minimal/r-minimal:3.6.3 bash
installr -p -c tidyverse/glue waldo

** byte-compile and prepare package for lazy loading
Error in get(genname, envir = envir) : object 'compare_proxy' not found
Calls: <Anonymous> ... register_S3_method_delayed -> registerS3method -> get

[hadley]

Hmmmm, interesting. I will try and figure out what makes compare_proxy different from the other method. Or maybe it's just the first loaded?

[gaborcsardi]

My (very uneducated) guess is that the difference is that waldo imports glue via NAMESPACE, whereas vctrs does not. So when waldo is being installed, glue is loaded, and that triggers the conditional generic, and R tries to load the generic, but since waldo is in a half-loaded state, it cannot.

So a simple workaround would be (if above is correct) to not import glue in NAMESPACE from waldo.

[hadley]

Do you have any easier way to test that? I made r-lib/waldo#195.

(This does feel like a bug in R to me, since it means you have to think more about how your packages interact, which our previous solution did not 😬)

[gaborcsardi]

Yep, your fix works. This is how to test it easily if you don't want to overwrite your system:

docker run -ti ghcr.io/r-hub/r-minimal/r-minimal:4.4.1 bash
installr -p -c tidyverse/glue r-lib/waldo@dont-import-glue

I agree that it is a bug in base R. A half-loaded package should not trigger the conditional S3 methods. (Although that'll make waldo's glue import behave differently at install time than at run time, but still better.)

[gaborcsardi]

Base R bug report: https://bugs.r-project.org/show_bug.cgi?id=18776

[jennybc]

We've encountered a variant of this problem before, with knitr.

See #253 and #254 (this PR especially contains a lot of write-up).