[R-pkg-devel] Installing "Additional" Packages During Tests

4 messages · brodie gaslam, Uwe Ligges

Original

1

4

Uwe Ligges

Mon, Mar 18, 2019 8:57 AM #

1. You should never install packages without asking the user.

2. a package should never write to the user file space or default 
libraries unless the user akss for it explicitly.
In tests, use tempdir().

3. You cannot expect that the library is writeable. It is not in many 
network installations.

Best,
Uwe Ligges

On 18.03.2019 16:48, brodie gaslam via R-package-devel wrote:

Subject:
Installing "Additional" Packages During Tests
From:
brodie gaslam <brodie.gaslam at yahoo.com>
Date:
18.03.2019, 16:48
To:
List R-package-devel <r-package-devel at r-project.org>

To:
List R-package-devel <r-package-devel at r-project.org>

Sorry, previous e-mail got pre-maturely sent due to fat finger...
My package unitizer[1] has recently gained the following type of error:
 ???? Warning in install.packages(pkg, repos = NULL, type = "src") :???? 'lib = "/home/hornik/tmp/R.check/r-devel-clang/Work/build/Packages"' is not writable
 ???? Error in install.packages(pkg, repos = NULL, type = "src") :
 ???? unable to install packages

While I did recently update the package to resolve the RNGversion issue, this problem seems unrelated as some of the CRAN machines produce it, and some don't, on the same package version.
The tests install some dummy packages with `install.packages`, and this has worked for a long time, but no longer, presumably due to changes in permissions of the test running daemon.? The packages are removed on.exit.? The packages are used for integration tests as unitizer is a package-testing-package and it is useful to be able to test functionality that includes use on actual packages that change.

Should I expect this to be the new normal, where I will be henceforth disallowed from installing packages temporarily?? Will there be an acceptable workaround (e.g. setting up a temporary library in a tempdir and try to `install.packages` into that; I admit I have no familiarity with this other than the vague awareness maybe this could be done)?
FWIW, perhaps an indication that this is something I shouldn't be doing came up earlier in the year when I had to resort to temporarily unsetting "R_TESTS" as per [2].
Thanks in advance for any input.
Best,
Brodie.

[1]: https://cran.r-project.org/web/checks/check_results_unitizer.html[2]: https://github.com/r-lib/testthat/issues/144
	[[alternative HTML version deleted]]

______________________________________________
R-package-devel at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-package-devel

Mon, Mar 18, 2019 9:28 AM #

I believe your first point makes the alternate workflow below illegal, but the subsequent ones add just enough ambiguity that I'd like to makes sure.? How about:
0. Only during the running of tests
1. Create a temporary directory tmplib using tempfile()/dircreate().
2. Install dummy package to that temporary directory using `lib=tmplib`: install.packages(srcdir, type='src', repos=NULL, lib=tmplib).
3. In the tests, use library(testpkg, lib.loc=tmplib)4. unload / remove / unlink temp directory on.exit
This is only in the test running.? None of the code in the package proper that the user would invoke in normal use installs packages.? It only happens if the user installs and runs the tests, and with the above modification, it would only be into a temporary library directory.
Thank you for your patience and attention.
Best,
Brodie.

On Monday, March 18, 2019, 11:56:55 AM EDT, Uwe Ligges <ligges at statistik.tu-dortmund.de> wrote:

1. You should never install packages without asking the user.

2. a package should never write to the user file space or default 
libraries unless the user akss for it explicitly.
In tests, use tempdir().

3. You cannot expect that the library is writeable. It is not in many 
network installations.

Best,
Uwe Ligges

On 18.03.2019 16:48, brodie gaslam via R-package-devel wrote:

Subject:
Installing "Additional" Packages During Tests
From:
brodie gaslam <brodie.gaslam at yahoo.com>
Date:
18.03.2019, 16:48
To:
List R-package-devel <r-package-devel at r-project.org>

To:
List R-package-devel <r-package-devel at r-project.org>

Sorry, previous e-mail got pre-maturely sent due to fat finger...
My package unitizer[1] has recently gained the following type of error:
? ???? Warning in install.packages(pkg, repos = NULL, type = "src") :???? 'lib = "/home/hornik/tmp/R.check/r-devel-clang/Work/build/Packages"' is not writable
? ???? Error in install.packages(pkg, repos = NULL, type = "src") :
? ???? unable to install packages

While I did recently update the package to resolve the RNGversion issue, this problem seems unrelated as some of the CRAN machines produce it, and some don't, on the same package version.
The tests install some dummy packages with `install.packages`, and this has worked for a long time, but no longer, presumably due to changes in permissions of the test running daemon.? The packages are removed on.exit.? The packages are used for integration tests as unitizer is a package-testing-package and it is useful to be able to test functionality that includes use on actual packages that change.

Should I expect this to be the new normal, where I will be henceforth disallowed from installing packages temporarily?? Will there be an acceptable workaround (e.g. setting up a temporary library in a tempdir and try to `install.packages` into that; I admit I have no familiarity with this other than the vague awareness maybe this could be done)?
FWIW, perhaps an indication that this is something I shouldn't be doing came up earlier in the year when I had to resort to temporarily unsetting "R_TESTS" as per [2].
Thanks in advance for any input.
Best,
Brodie.

[1]: https://cran.r-project.org/web/checks/check_results_unitizer.html[2]: https://github.com/r-lib/testthat/issues/144
??? [[alternative HTML version deleted]]

______________________________________________
R-package-devel at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-package-devel

Uwe Ligges

Mon, Mar 18, 2019 9:34 AM #

Simply resubmit such a package, we will lookninto the details and the 
team will decide.

Best,
Uwe  Ligges

On 18.03.2019 17:28, brodie gaslam wrote:

I believe your first point makes the alternate workflow below illegal, 
but the subsequent ones add just enough ambiguity that I'd like to makes 
sure.? How about:

0. Only during the running of tests
1. Create a temporary directory tmplib using tempfile()/dircreate().
2. Install dummy package to that temporary directory using `lib=tmplib`: 
install.packages(srcdir, type='src', repos=NULL, lib=tmplib).
3. In the tests, use library(testpkg, lib.loc=tmplib)
4. unload / remove / unlink temp directory on.exit

This is only in the test running.? None of the code in the package 
proper that the user would invoke in normal use installs packages.? It 
only happens if the user installs and runs the tests, and with the above 
modification, it would only be into a temporary library directory.

Thank you for your patience and attention.

Best,

Brodie.


On Monday, March 18, 2019, 11:56:55 AM EDT, Uwe Ligges 
<ligges at statistik.tu-dortmund.de> wrote:


1. You should never install packages without asking the user.

2. a package should never write to the user file space or default
libraries unless the user akss for it explicitly.
In tests, use tempdir().

3. You cannot expect that the library is writeable. It is not in many
network installations.

Best,
Uwe Ligges



On 18.03.2019 16:48, brodie gaslam via R-package-devel wrote:

 >
 > Subject:
 > Installing "Additional" Packages During Tests
 > From:
 > brodie gaslam <brodie.gaslam at yahoo.com <mailto:brodie.gaslam at yahoo.com>>
 > Date:
 > 18.03.2019, 16:48
 > To:
 > List R-package-devel <r-package-devel at r-project.org

<mailto:r-package-devel at r-project.org>>

 >
 > To:
 > List R-package-devel <r-package-devel at r-project.org

<mailto:r-package-devel at r-project.org>>

 >
 >
 > Sorry, previous e-mail got pre-maturely sent due to fat finger...
 > My package unitizer[1] has recently gained the following type of error:
 >? ???? Warning in install.packages(pkg, repos = NULL, type = "src")

:???? 'lib = 
"/home/hornik/tmp/R.check/r-devel-clang/Work/build/Packages"' is not 
writable

 >? ???? Error in install.packages(pkg, repos = NULL, type = "src") :
 >? ???? unable to install packages
 >
 > While I did recently update the package to resolve the RNGversion

issue, this problem seems unrelated as some of the CRAN machines produce 
it, and some don't, on the same package version.

 > The tests install some dummy packages with `install.packages`, and

this has worked for a long time, but no longer, presumably due to 
changes in permissions of the test running daemon.? The packages are 
removed on.exit.? The packages are used for integration tests as 
unitizer is a package-testing-package and it is useful to be able to 
test functionality that includes use on actual packages that change.

 >
 > Should I expect this to be the new normal, where I will be henceforth

disallowed from installing packages temporarily?? Will there be an 
acceptable workaround (e.g. setting up a temporary library in a tempdir 
and try to `install.packages` into that; I admit I have no familiarity 
with this other than the vague awareness maybe this could be done)?

 > FWIW, perhaps an indication that this is something I shouldn't be

doing came up earlier in the year when I had to resort to temporarily 
unsetting "R_TESTS" as per [2].

 > Thanks in advance for any input.
 > Best,
 > Brodie.
 >
 >
 > [1]:

https://cran.r-project.org/web/checks/check_results_unitizer.html[2]: 
https://github.com/r-lib/testthat/issues/144

 > ??? [[alternative HTML version deleted]]
 >
 >
 > ______________________________________________
 > R-package-devel at r-project.org <mailto:R-package-devel at r-project.org>

mailing list

 > https://stat.ethz.ch/mailman/listinfo/r-package-devel
 >

Uwe Ligges

Mon, Mar 18, 2019 9:36 AM #

On 18.03.2019 17:28, brodie gaslam wrote:

Please submit following these suggestions and the CRAN team will look 
into the details and decide.

Best,
Uwe

Best,

Brodie.


On Monday, March 18, 2019, 11:56:55 AM EDT, Uwe Ligges 
<ligges at statistik.tu-dortmund.de> wrote:


1. You should never install packages without asking the user.

2. a package should never write to the user file space or default
libraries unless the user akss for it explicitly.
In tests, use tempdir().

3. You cannot expect that the library is writeable. It is not in many
network installations.

Best,
Uwe Ligges



On 18.03.2019 16:48, brodie gaslam via R-package-devel wrote:

 >
 > Subject:
 > Installing "Additional" Packages During Tests
 > From:
 > brodie gaslam <brodie.gaslam at yahoo.com <mailto:brodie.gaslam at yahoo.com>>
 > Date:
 > 18.03.2019, 16:48
 > To:
 > List R-package-devel <r-package-devel at r-project.org

<mailto:r-package-devel at r-project.org>>

 >
 > To:
 > List R-package-devel <r-package-devel at r-project.org

<mailto:r-package-devel at r-project.org>>

 >
 >
 > Sorry, previous e-mail got pre-maturely sent due to fat finger...
 > My package unitizer[1] has recently gained the following type of error:
 >? ???? Warning in install.packages(pkg, repos = NULL, type = "src")

:???? 'lib = 
"/home/hornik/tmp/R.check/r-devel-clang/Work/build/Packages"' is not 
writable

 >? ???? Error in install.packages(pkg, repos = NULL, type = "src") :
 >? ???? unable to install packages
 >
 > While I did recently update the package to resolve the RNGversion

issue, this problem seems unrelated as some of the CRAN machines produce 
it, and some don't, on the same package version.

 > The tests install some dummy packages with `install.packages`, and

this has worked for a long time, but no longer, presumably due to 
changes in permissions of the test running daemon.? The packages are 
removed on.exit.? The packages are used for integration tests as 
unitizer is a package-testing-package and it is useful to be able to 
test functionality that includes use on actual packages that change.

 >
 > Should I expect this to be the new normal, where I will be henceforth

disallowed from installing packages temporarily?? Will there be an 
acceptable workaround (e.g. setting up a temporary library in a tempdir 
and try to `install.packages` into that; I admit I have no familiarity 
with this other than the vague awareness maybe this could be done)?

 > FWIW, perhaps an indication that this is something I shouldn't be

doing came up earlier in the year when I had to resort to temporarily 
unsetting "R_TESTS" as per [2].

 > Thanks in advance for any input.
 > Best,
 > Brodie.
 >
 >
 > [1]:

https://cran.r-project.org/web/checks/check_results_unitizer.html[2]: 
https://github.com/r-lib/testthat/issues/144

 > ??? [[alternative HTML version deleted]]
 >
 >
 > ______________________________________________
 > R-package-devel at r-project.org <mailto:R-package-devel at r-project.org>

mailing list

 > https://stat.ethz.ch/mailman/listinfo/r-package-devel
 >