[R-SIG-Mac] R CMD check "HTML version of manual" NOTE/Warnings

[John Fox]

Dear R-sig-mac list members,

When checking packages --as-cran with R 4.2.0 (and R 4.2.0 patched), I'm
seeing multiple warnings (and a NOTE) concerning the HTML version of the
package help-page manuals. The warning appears for every .Rd file in all
of the packages that I've checked. I didn't see this problem before R
4.2.0.

I'm writing to the r-sig-mac list rather than r-package-devel because I
don't encounter the same problem under Windows. Nor does it appear on
the CRAN check pages for the packages.

I know that there's been discussion of adding an HTML manual, which
seems a good idea, but I didn't realize that this has apparently already
been implemented.

Has anyone else experienced this problem or does anyone understand its
source? AFAIK, there's nothing unusual about the R installation on my
Mac, but of course there may be some setting that inadvertently turned
on checking the HTML manual.

Here's an example (with many lines elided, . . .):

------ snip ------

Johns-MacBook-Pro:car johnfox$ R CMD check --as-cran car_3.0-13.tar.gz
* using log directory
'/Users/johnfox/Documents/R-package-sources/car/car.Rcheck'
* using R version 4.2.0 Patched (2022-04-24 r82246)
* using platform: aarch64-apple-darwin20 (64-bit)
* using session charset: UTF-8
* using option '--as-cran'
* checking for file 'car/DESCRIPTION' ... OK
* this is package 'car' version '3.0-13'
* checking CRAN incoming feasibility ... Note_to_CRAN_maintainers
Maintainer: 'John Fox <jf...@mcmaster.ca>'
* checking package namespace information ... OK

. . .

* checking PDF version of manual ... OK
* checking HTML version of manual ... NOTE
Found the following problems:
Anova.Rd:4:1: Warning: <link> inserting "type" attribute
Anova.Rd:12:1: Warning: <script> proprietary attribute "onload"
Anova.Rd:12:1: Warning: <script> inserting "type" attribute
Anova.Rd:17:1: Warning: <table> lacks "summary" attribute
Anova.Rd:135:1: Warning: <table> lacks "summary" attribute
Anova.Rd:285:1: Warning: <table> lacks "summary" attribute
Boot.Rd:4:1: Warning: <link> inserting "type" attribute
Boot.Rd:12:1: Warning: <script> proprietary attribute "onload"
Boot.Rd:12:1: Warning: <script> inserting "type" attribute
Boot.Rd:17:1: Warning: <table> lacks "summary" attribute
Boot.Rd:56:1: Warning: <table> lacks "summary" attribute

. . .

which.names.Rd:12:1: Warning: <script> proprietary attribute "onload"
which.names.Rd:12:1: Warning: <script> inserting "type" attribute
which.names.Rd:17:1: Warning: <table> lacks "summary" attribute
which.names.Rd:44:1: Warning: <table> lacks "summary" attribute
* checking for non-standard things in the check directory ... OK
* checking for detritus in the temp directory ... OK
* DONE

Status: 1 NOTE
See
'/Users/johnfox/Documents/R-package-sources/car/car.Rcheck/00check.log'
for details.

------ snip ------

Any help would be appreciated.

Thanks,
John
--
John Fox, Professor Emeritus
McMaster University
Hamilton, Ontario, Canada
web: https://socialsciences.mcmaster.ca/jfox/

_______________________________________________
R-SIG-Mac mailing list
R-SI...@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-mac

[Prof Brian Ripley]

On 25/04/2022 20:04, John Fox wrote:
> Dear R-sig-mac list members,
>
> When checking packages --as-cran with R 4.2.0 (and R 4.2.0 patched), I'm
> seeing multiple warnings (and a NOTE) concerning the HTML version of the
> package help-page manuals. The warning appears for every .Rd file in all
> of the packages that I've checked. I didn't see this problem before R
> 4.2.0.
>
> I'm writing to the r-sig-mac list rather than r-package-devel because I

Unfortunately the authors of this are not on r-sig-mac.

> don't encounter the same problem under Windows. Nor does it appear on
> the CRAN check pages for the packages.

This check is only done if you have 'tidy' on the path. My Monterey M1
MBP has

auk2% tidy --version
HTML Tidy for Mac OS X released on 31 October 2006 - Apple Inc. build 2649

I think that is far too old. The short answer is to ignore these, or
update tidy (from http://binaries.html-tidy.org/) which (5.8.0) finds
different issues for car that I do not see on Fedora (with 5.7.16).

Probably your Windows machine does not have tidy installed.

> I know that there's been discussion of adding an HTML manual, which
> seems a good idea, but I didn't realize that this has apparently already
> been implemented.
>
> Has anyone else experienced this problem or does anyone understand its
> source? AFAIK, there's nothing unusual about the R installation on my
> Mac, but of course there may be some setting that inadvertently turned
> on checking the HTML manual.

--as-cran turned it on ((f tidy is available).

Brian D. Ripley, rip...@stats.ox.ac.uk
Emeritus Professor of Applied Statistics, University of Oxford

[John Fox]

Dear Brian,

Thank you very much for the explanation. Please see interline below:

On 2022-04-25 4:51 p.m., Prof Brian Ripley wrote:
> On 25/04/2022 20:04, John Fox wrote:
>> Dear R-sig-mac list members,
>>
>> When checking packages --as-cran with R 4.2.0 (and R 4.2.0 patched),
>> I'm seeing multiple warnings (and a NOTE) concerning the HTML version
>> of the package help-page manuals. The warning appears for every .Rd
>> file in all of the packages that I've checked. I didn't see this
>> problem before R 4.2.0.
>>
>> I'm writing to the r-sig-mac list rather than r-package-devel because I
>
> Unfortunately the authors of this are not on r-sig-mac.

Should I mention this as well on r-package-devel?

>
>> don't encounter the same problem under Windows. Nor does it appear on
>> the CRAN check pages for the packages.
>
> This check is only done if you have 'tidy' on the path.  My Monterey M1
> MBP has
>
> auk2% tidy --version
> HTML Tidy for Mac OS X released on 31 October 2006 - Apple Inc. build 2649

That's exactly the version installed on my MacBook Pro. Is this standard
for Monterey? If not, would it be best simply to remove tidy (and how
did it get there?).

>
> I think that is far too old.  The short answer is to ignore these, or
> update tidy (from http://binaries.html-tidy.org/) which (5.8.0) finds
> different issues for car that I do not see on Fedora (with 5.7.16).

I saw these warnings for every .Rd file in all of the packages I
checked, not just for the car package. The HTML is presumably generated
from the .Rd files by the R installation process. If tidy is general, or
common, on Monterey Macs, I wonder why others aren't reporting similar
warnings on R CMD check --as-cran.

>
> Probably your Windows machine does not have tidy installed.
>
>> I know that there's been discussion of adding an HTML manual, which
>> seems a good idea, but I didn't realize that this has apparently
>> already been implemented.
>>
>> Has anyone else experienced this problem or does anyone understand its
>> source? AFAIK, there's nothing unusual about the R installation on my
>> Mac, but of course there may be some setting that inadvertently turned
>> on checking the HTML manual.
>
> --as-cran turned it on ((f tidy is available).

Again, thank you for the explanation.

Best,
John

John Fox, Professor Emeritus
McMaster University
Hamilton, Ontario, Canada
web: https://socialsciences.mcmaster.ca/jfox/

[peter dalgaard]

Yes, I think this should go to r-package-devel or maybe even file a bug report against "R CMD check". Looks like we "just" need to revise the logic at the start of check_Rd2HTML():

check_Rd2HTML <- function(pkgdir) {
if(!nzchar(Sys.which("tidy")) ||
!startsWith(system2("tidy", "--version", stdout = TRUE),
"HTML Tidy"))
return()

to either

- look for a version number (which the Apple builds don't have)
- explicitly reject the 2006 Apple builds
- somehow allow for an override using an environment variable

or thereabouts...

- pd

--
Peter Dalgaard, Professor,
Center for Statistics, Copenhagen Business School
Solbjerg Plads 3, 2000 Frederiksberg, Denmark
Phone: (+45)38153501
Office: A 4.23
Email: pd....@cbs.dk Priv: PDa...@gmail.com

[Prof Brian Ripley]

On 26/04/2022 15:43, peter dalgaard wrote:
> Yes, I think this should go to r-package-devel or maybe even file a bug report against "R CMD check". Looks like we "just" need to revise the logic at the start of check_Rd2HTML():
>
> check_Rd2HTML <- function(pkgdir) {
> if(!nzchar(Sys.which("tidy")) ||
> !startsWith(system2("tidy", "--version", stdout = TRUE),
> "HTML Tidy"))
> return()
>
> to either
>
> - look for a version number (which the Apple builds don't have)
> - explicitly reject the 2006 Apple builds
> - somehow allow for an override using an environment variable

This is switched on by _R_CHECK_RD_VALIDATE_RD2HTML_=true or --as-cran.
I plan to at least let _R_CHECK_RD_VALIDATE_RD2HTML_=false override
--as-cran.

There seems to have been a report on r-package-devel not mentioning macOS:

https://stat.ethz.ch/pipermail/r-package-devel/2022q1/007804.html
https://stat.ethz.ch/pipermail/r-package-devel/2022q2/007932.html

but I don't think anyone had a clue that the macOS 'tidy' might be
insufficient.

Brian D. Ripley, rip...@stats.ox.ac.uk
Emeritus Professor of Applied Statistics, University of Oxford

[John Fox]

Dear Brian and Peter,

As Brian mentioned, I still get warnings (though different ones) from
the updated tidy. For example, for the car package:

------ snip -------

* checking HTML version of manual ... NOTE
Found the following problems:
Anova.Rd:4:1: Warning: <link> inserting "type" attribute
Anova.Rd:12:1: Warning: <script> proprietary attribute "onload"
Anova.Rd:12:1: Warning: <script> inserting "type" attribute
Anova.Rd:17:1: Warning: <table> lacks "summary" attribute
Anova.Rd:135:1: Warning: <table> lacks "summary" attribute
Anova.Rd:285:1: Warning: <table> lacks "summary" attribute

------ snip -------

As before, these warnings are generated for all of the .Rd files in the
package.

I'm cc'ing the r-package-devel list (with apologies for cross-posting)
so that there will be a record there.

Best,
John

[Prof Brian Ripley]

To round this off, R-patched has the NEWS item

• Setting environment variable _R_CHECK_RD_VALIDATE_RD2HTML_ to a
false value will override R CMD check --as-cran and turn off HTML
validation. This provides a way to circumvent a problematic
tidy.

The 2006 version that ships with macOS is always skipped.

and R-admin now says

R CMD check --as-cran also makes use of ‘HTML Tidy’. macOS has a version
in /usr/bin/tidy dating from 2006 which is far too old and is skipped.
Up-to-date versions can be installed from http://binaries.html-tidy.org/.

The current version there is 5.8.0, later than Debian's 5.6.0 used in
developing this and detecting further issues in Rd2HTML that have since
been addressed.

So the simplest thing for someone wanting to do an -as-cran check is to
install html-tidy 5.8.0.

Unfortunately there is nothing we can do about 4.2.0.