warning: no visible global function definition in continuous integration

[ElsLommelen]

Today I got a lot of warnings after running lintr during continuous integration on Wercker, which are not related to the few changes I committed (and yesterday CI passed):
Warning: no visible global function definition for '...'

The warnings mention either a function of the package itself or an imported function, which I added under imports in the description file, and after @importFrom in the Roxygen2 documentation (and these functions were added to the namespace file).

I tried running lintr::lint.package() locally in the project folder of the package in RStudio, but this did not give any warnings.

Here is the commit and here the report with warnings on Wercker. Any idea what could be wrong?

[russHyde]

Does it pass R CMD check locally?

[ElsLommelen]

Yes, it passes R CMD check.

[russHyde]

It looks like a server problem:

Bad Gateway (HTTP 502). Failed to {
  "message": "Server Error"
}

If it is reproducible, then it might be an issue with https://github.com/jimhester/wercker-step-r-lint rather than lintr itself.

[ElsLommelen]

Honestly I have no idea how to check if this is a server problem.
The warnings showed up again after each new commit. For now I decided to ignore the object_usage_linter temporary to be able to continue working.
(To test reproducibility, I started dublicating the repo, but it turned out to be much more work than I expected. I will continu next week.)

[russHyde]

Sorry, I think the server problem in your wercker run was unrelated to the issue you've raised.
I've just cloned LSVI and ran lintr on it, and have received the object-usage lints.

[ElsLommelen]

I started a new repo in which I already simplified the package a bit: test_lintr_on_wercker. I have not yet pinpointed the problem, but I already learnt that it is probably OS specific to Unix systems:

• On a windows computer no warning is given locally in R CMD check or lintr::lint_package() (commits of 13 march)
• On a Linux computer, on the other hand, warnings may also show up locally in R CMD check or lintr::lint_package(). This was the case in this commit, but due to solving other problems, the local check did not give the specific warnings before doing this commit (but it failed on wercker in a rocker/verse container anyway).

[ElsLommelen]

I simplified the repo test_lintr_on_wercker till I could pinpoint the problem: if a SQLite database and/or a connection to a SQLite database is present in the package, the warning in lintr shows up on a Unix OS for all functions preceded by a line of comments in the previous line.

The one but last commit seems the most simplified case:

• the function selecteerIndicatoren has a line of comments and 2 imported functions that evoke a warning
• the package moreover contains a SQLite database and a function connecteerMetLSVIdb that makes a connection with this database (not used in ´selecteerIndicatoren´)

After removing the SQLite database and related functions in the last commit, the warnings disappeared.

Previous commits show that the warning is given only when the comment is on the line preceding the funtion:

• in this commit I added a comment on line 23 (and simplified the function) and I noticed the function on line 24 gave a warning, which was not the case in the previous commit
• in this commit a new warning shows up after moving dbGetQuery() from line 50 (2 lines after comment) to line 49 (line after comment)

There is one important question that I could not solve and that might lead to a solution: why did the warnings suddenly appear on the 10th of mars? In the package itself there were no important changes that could explain their appearance. That day no new version of the RSQLite package was published.

(In the package I switched to English for relevant information and comments on this issue. So just ignore the Dutch parts; for my own convenience I did not change the function names but I think the above information is sufficient to understand it?)

I hope this helps to solve the problem? Admittedly I still do not know in which package this issue should be solved.

[ElsLommelen]

In this package (which I hadn't looked for for years before I did some changes yesterday) I encounter the same problem (which first occurred in PR 28), but in a slightly different setting:

• the main package with functions and documentation does not contain a database, but the inst-folder contains a script hoofdscript.R to retain data from an Access database (which is put there as an example for a limited group of users)
• here we use package RODBC to connect to the database, whereas we used DBI and odbc in the previously reported package

Admittedly it is an unusual situation to have a script with an Access database tested in a Linux environment, but I hope it can help to tackle the problem. For now I will ignore the object_usage_linter for my own convenience.

[AshesITR]

Can you check if this was fixed by #740?
Additionally, I found it advisable to devtools::load_all() the to-be-linted package as well before running lint_package() - this ensures all functions and objects from all source files are visible to checkUsage().

[ElsLommelen]

I checked both of my previously reported cases, and the warnings did not show up anymore.
Thanks for fixing this!

[AshesITR]

That's good to hear! Thanks for checking.