[tex-live] inclusion of texlive-includernw

[Roger Bivand]

The inclusion in current texlive of texlive-includernw (depending in
packaged systems on R-knitr, not a texlive component), and its listing in
texlive-collection-mathscience, hence in texlive-scheme-medium, is causing
havoc for those who upgrade texlive because texlive-includernw tries to
install R and many R packages without checking whether they are installed
elsewhere. I can't see a list of maintainers of
texlive-collection-mathscience, so cannot see how to get
texlive-includernw de-listed urgently. It should be free-standing only,
and should be re-written to check for dependency availability on R at the
level of the user rather than root at install time. This probably cannot
be done. The package should actually do the checking at run time only,
providing helpful messages only.

The problem may be an over-ambitious Texlive Fedora packager - the issue
hit me on upgrading from Fedora 28 to Fedora 29 today.

Roger Bivand

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: [phone]; e-mail: ***@nhh.no
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en

[Andreas Storvik Strauman]

> causing havoc for those who upgrade texlive because texlive-includernw tries to
> install R and many R packages without checking whether they are installed
> elsewhere.


Oh! I see. That's a problem.


> I can't see a list of maintainers of texlive-collection-mathscience, so cannot see how to get texlive-includernw de-listed urgently.


I don't know who put it there! I do not know how it ended up in TeXLive fedora,

but I do have email addresses of some TeXLive guys, and I could try to contact them?


> ... should be re-written to check for dependency availability on R at the
> level of the user rather than root at install time.

Agreed.

> This probably cannot be done.

Agreed, but I'll give it a shot anyways.


>The package should actually do the checking at run time only,

> providing helpful messages only.


At run time is better than install time, or no?


I did not intend this package to cause any inconvenience, and

I find it extremely odd that it's actually attempting to install R-knitr?


________________________________
From: Roger Bivand <***@nhh.no>
Sent: Saturday, November 10, 2018 6:59:52 PM
To: tex-***@tug.org
Cc: Andreas Storvik Strauman
Subject: inclusion of texlive-includernw

The inclusion in current texlive of texlive-includernw (depending in
packaged systems on R-knitr, not a texlive component), and its listing in
texlive-collection-mathscience, hence in texlive-scheme-medium, is causing
havoc for those who upgrade texlive because texlive-includernw tries to
install R and many R packages without checking whether they are installed
elsewhere. I can't see a list of maintainers of
texlive-collection-mathscience, so cannot see how to get
texlive-includernw de-listed urgently. It should be free-standing only,
and should be re-written to check for dependency availability on R at the
level of the user rather than root at install time. This probably cannot
be done. The package should actually do the checking at run time only,
providing helpful messages only.

The problem may be an over-ambitious Texlive Fedora packager - the issue
hit me on upgrading from Fedora 28 to Fedora 29 today.

Roger Bivand

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: [phone]; e-mail: ***@nhh.no
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en

[Roger Bivand]

On Sat, 10 Nov 2018, Andreas Storvik Strauman wrote:

>> causing havoc for those who upgrade texlive because texlive-includernw tries to
>> install R and many R packages without checking whether they are installed
>> elsewhere.
>
>
> Oh! I see. That's a problem.
>
>
>> I can't see a list of maintainers of texlive-collection-mathscience, so cannot see how to get texlive-includernw de-listed urgently.
>
>
> I don't know who put it there! I do not know how it ended up in TeXLive
> fedora, but I do have email addresses of some TeXLive guys, and I could
> try to contact them?

Excellent, and congratulations on being included - it means others find
your contribution useful! But please do try to investigate whether it is
the inclusion in the collection (I believe this is the case) that causes
the install-time checks, or whether it is a mis-understanding by the
Fedora texlive packager.

>
>
>> ... should be re-written to check for dependency availability on R at the
>> level of the user rather than root at install time.
>
> Agreed.
>
>> This probably cannot be done.
>
> Agreed, but I'll give it a shot anyways.
>
>
>> The package should actually do the checking at run time only,
>
>> providing helpful messages only.
>
>
> At run time is better than install time, or no?
>
>
> I did not intend this package to cause any inconvenience, and I find it
> extremely odd that it's actually attempting to install R-knitr?
>

That's what it does on Fedora. Maybe there are specification settings in
the package (is the source on a repo, if so where?), but it could be the
Fedora texlive packager.

Sorry to jump on you, but this cost me a good deal of worry before I
understood why upgrading Fedora had installed an R rpm with R packages
including R-knitr.

Roger

>
> ________________________________
> From: Roger Bivand <***@nhh.no>
> Sent: Saturday, November 10, 2018 6:59:52 PM
> To: tex-***@tug.org
> Cc: Andreas Storvik Strauman
> Subject: inclusion of texlive-includernw
>
> The inclusion in current texlive of texlive-includernw (depending in
> packaged systems on R-knitr, not a texlive component), and its listing in
> texlive-collection-mathscience, hence in texlive-scheme-medium, is causing
> havoc for those who upgrade texlive because texlive-includernw tries to
> install R and many R packages without checking whether they are installed
> elsewhere. I can't see a list of maintainers of
> texlive-collection-mathscience, so cannot see how to get
> texlive-includernw de-listed urgently. It should be free-standing only,
> and should be re-written to check for dependency availability on R at the
> level of the user rather than root at install time. This probably cannot
> be done. The package should actually do the checking at run time only,
> providing helpful messages only.
>
> The problem may be an over-ambitious Texlive Fedora packager - the issue
> hit me on upgrading from Fedora 28 to Fedora 29 today.
>
> Roger Bivand
>
> --
> Roger Bivand
> Department of Economics, Norwegian School of Economics,
> Helleveien 30, N-5045 Bergen, Norway.
> voice: [phone]; e-mail: ***@nhh.no
> http://orcid.org/0000-0003-2392-6140
> https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en
>

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: [phone]; e-mail: ***@nhh.no
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en

[Karl Berry]

This list (tex-***@tug.org) is about upstream TeX Live. We have no
influence on what Fedora, or any distro, does wrt dependencies or
anything else. I suggest submitting a bug in whatever way Fedora
prefers. --good luck, karl.

[Roger Bivand]

On Sat, 10 Nov 2018, Karl Berry wrote:

> This list (tex-***@tug.org) is about upstream TeX Live. We have no
> influence on what Fedora, or any distro, does wrt dependencies or
> anything else. I suggest submitting a bug in whatever way Fedora
> prefers. --good luck, karl.

Thanks, I understand, but also feel that the includernw latex package
*.sty should itself test for a working R environment, and whether that
environment includes the R knitr package, stopping if they are not found
and issuing helpful instructions to the user. The Fedora packager
might then have been less eager to lumber all texlive-scheme-medium
users with a large and unnecessary dependency. Last year I was editor in
chief of the R Journal (responsible for technical production of articles),
and need precise control over workflow components in this capacity. The
invasiveness of this upgrade is really shocking.

Roger

>
>

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: [phone]; e-mail: ***@nhh.no
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en

[Norbert Preining]

Hi,

as Karl said, this seems to be a Fedora only issue.

Furthermore, I am suprised about this:

> havoc for those who upgrade texlive because texlive-includernw tries to
> install R and many R packages without checking whether they are installed
> elsewhere. I can't see a list of maintainers of

Do you mean *using* the .sty file and running a tex file using it
installs R packages? I looked into the sty file and only saw R code
being executed of the form
library('knitr')
..
On my system this does *not* automatically install anything, but gives
an error if knitr is not installed.

OTOH, if the *Fedora package* (rpm) depends on R packages, then this is
the correct way to satisfy dependencies. This is how *every* package
manager works.

Thanks

Norbert

--
PREINING Norbert http://www.preining.info
Accelia Inc. + JAIST + TeX Live + Debian Developer
GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13

[Roger Bivand]

On Sun, 11 Nov 2018, Norbert Preining wrote:

> Hi,
>
> as Karl said, this seems to be a Fedora only issue.
>
> Furthermore, I am suprised about this:
>
>> havoc for those who upgrade texlive because texlive-includernw tries to
>> install R and many R packages without checking whether they are installed
>> elsewhere. I can't see a list of maintainers of
>
> Do you mean *using* the .sty file and running a tex file using it
> installs R packages? I looked into the sty file and only saw R code
> being executed of the form
> library('knitr')
> ..
> On my system this does *not* automatically install anything, but gives
> an error if knitr is not installed.
>
> OTOH, if the *Fedora package* (rpm) depends on R packages, then this is
> the correct way to satisfy dependencies. This is how *every* package
> manager works.

No, not every. The R package manager (so the R code in this latex package)
advises limiting the number of dependencies (Depends, Imports) as far as
possible, and moving everything else into Suggests, which are checked at
run time, not install time. A user then finding an unfulfilled dependency
at run time is at liberty to install the missing workflow components, but
is never left without a choice by the packaging system.

So "correct" has many flavours. I will try to trace the Fedora packager,
and follow up with the author of includernw, who certainly did not ask for
the latex package to be rolled out at least to all texlive-scheme-medium
users on Fedora.

Roger

>
> Thanks
>
> Norbert
>
> --
> PREINING Norbert http://www.preining.info
> Accelia Inc. + JAIST + TeX Live + Debian Developer
> GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13
>

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: [phone]; e-mail: ***@nhh.no
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en

[Norbert Preining]

Hi Roger,

I have read now both your answers, but still tap in the dark with what
you want to achieve and what is your actual problem. Your descriptions
are a complete mixture of nomenclature.

On Sun, 11 Nov 2018, Roger Bivand wrote:
> No, not every. The R package manager (so the R code in this latex package)
> advises limiting the number of dependencies (Depends, Imports) as far as

I don't know what you are talking about, honestly.

The latex package you are refering to does *NOT* contain any package
manager facilities. I calls R on some code which calls
"library('knitr')".

You are mixing completely different things here: depends, suggests etc
are entities of the fedora packaging, and have nothing to do with the
latex package.

Upstream (other of includenrw) has no influence whatsoever what fedora
packager are doing.

> So "correct" has many flavours. I will try to trace the Fedora packager, and
> follow up with the author of includernw, who certainly did not ask for the
> latex package to be rolled out at least to all texlive-scheme-medium users
> on Fedora.

Again wrong. includenrw is included in TeX Live. And the Fedora
maintainer has packaged TeX Live and thus also includenrw.
There is nothing wrong, and if at all, then it is in your handling of
how you install R packages and manage them.

It is *you* who has to decide whether you want to use the fedora
packaged versions of R pakcages, or locally isntalled ones.

BTW, I know what I am talking about: I am TeX Live developer, at the
same time maintainer of TeX Live in Debian, and do quite a lot of work
with R. So trust me, it is in *your* obligation to keep a workable
environment.

If you start installing local R packages you take the control intio your
hands, but also the responsability to have all the necessary tools etc
installed.

Don't blame either the author of includenrw or the fedora maintainer for
problems.

Best

Norbert

--
PREINING Norbert http://www.preining.info
Accelia Inc. + JAIST + TeX Live + Debian Developer
GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13

[Andreas Storvik Strauman]

As I have come to understand it, the problem is that Fedora

installs R together with texlive-collection-mathscience because of

includernw.sty, due to it's dependency on R and R-knitr.


However, Roger does not want to use includernw, and has R installed

(in a non-conventional location on his computer). This leaves Roger

with two R installations and a .sty he doesn't use.


Am I correct, Roger?

________________________________
From: Norbert Preining <***@logic.at>
Sent: Sunday, November 11, 2018 3:14:29 PM
To: Roger Bivand
Cc: Andreas Storvik Strauman; tex-***@tug.org
Subject: Re: [tex-live] inclusion of texlive-includernw

Hi Roger,

I have read now both your answers, but still tap in the dark with what
you want to achieve and what is your actual problem. Your descriptions
are a complete mixture of nomenclature.

On Sun, 11 Nov 2018, Roger Bivand wrote:
> No, not every. The R package manager (so the R code in this latex package)
> advises limiting the number of dependencies (Depends, Imports) as far as

I don't know what you are talking about, honestly.

The latex package you are refering to does *NOT* contain any package
manager facilities. I calls R on some code which calls
"library('knitr')".

You are mixing completely different things here: depends, suggests etc
are entities of the fedora packaging, and have nothing to do with the
latex package.

Upstream (other of includenrw) has no influence whatsoever what fedora
packager are doing.

> So "correct" has many flavours. I will try to trace the Fedora packager, and
> follow up with the author of includernw, who certainly did not ask for the
> latex package to be rolled out at least to all texlive-scheme-medium users
> on Fedora.

Again wrong. includenrw is included in TeX Live. And the Fedora
maintainer has packaged TeX Live and thus also includenrw.
There is nothing wrong, and if at all, then it is in your handling of
how you install R packages and manage them.

It is *you* who has to decide whether you want to use the fedora
packaged versions of R pakcages, or locally isntalled ones.

BTW, I know what I am talking about: I am TeX Live developer, at the
same time maintainer of TeX Live in Debian, and do quite a lot of work
with R. So trust me, it is in *your* obligation to keep a workable
environment.

If you start installing local R packages you take the control intio your
hands, but also the responsability to have all the necessary tools etc
installed.

Don't blame either the author of includenrw or the fedora maintainer for
problems.

Best

Norbert

--
PREINING Norbert http://www.preining.info
[http://www.preining.info/front-page.png]<http://www.preining.info/>

Norbert Preining<http://www.preining.info/>
www.preining.info
latest galleries: (c) 2012-2018 Norbert Preining



Accelia Inc. + JAIST + TeX Live + Debian Developer
GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13

[Norbert Preining]

Hi

> As I have come to understand it, the problem is that Fedora
> installs R together with texlive-collection-mathscience because of
> includernw.sty, due to it's dependency on R and R-knitr.

Yeah, this is what I expect.

> However, Roger does not want to use includernw, and has R installed
> (in a non-conventional location on his computer). This leaves Roger
> with two R installations and a .sty he doesn't use.

And? Is disk space the problem? If Roger has his installation say in
/opt/R/bin
/opt/R/lib
and the Fedora packages will probably install into
/usr/bin/R
/usr/lib/R/...
then, if the version in /opt/R/bin is in the front of the PATH, it will
be used and will not - unless explicitly told to - use the packages of
Fedora in /usr/lib/R

This is the very same case with TeX installations by the packaging
system and the original from TUG. Setting up the PATH correctly solves
these problems.

If it is the disk space, then Roger can take a similar approach as
explained for Debian and TeX Live in
https://www.tug.org/texlive/debian.html
under
Integrating vanilla TeX Live with Debian
one needs to create a dummy packages the tells the package manager (rpm
in Roger's case) that there is a full R installation available.

But this is only necessary when disk space is scarce, what in the case
of R should make a huge difference.

Best

Norbert

--
PREINING Norbert http://www.preining.info
Accelia Inc. + JAIST + TeX Live + Debian Developer
GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13

[Roger Bivand]

On Sun, 11 Nov 2018, Norbert Preining wrote:

> Hi
>
>> As I have come to understand it, the problem is that Fedora
>> installs R together with texlive-collection-mathscience because of
>> includernw.sty, due to it's dependency on R and R-knitr.
>
> Yeah, this is what I expect.
>
>> However, Roger does not want to use includernw, and has R installed
>> (in a non-conventional location on his computer). This leaves Roger
>> with two R installations and a .sty he doesn't use.
>
> And? Is disk space the problem? If Roger has his installation say in
> /opt/R/bin
> /opt/R/lib
> and the Fedora packages will probably install into
> /usr/bin/R
> /usr/lib/R/...
> then, if the version in /opt/R/bin is in the front of the PATH, it will
> be used and will not - unless explicitly told to - use the packages of
> Fedora in /usr/lib/R
>
> This is the very same case with TeX installations by the packaging
> system and the original from TUG. Setting up the PATH correctly solves
> these problems.
>
> If it is the disk space, then Roger can take a similar approach as
> explained for Debian and TeX Live in
> https://www.tug.org/texlive/debian.html
> under
> Integrating vanilla TeX Live with Debian
> one needs to create a dummy packages the tells the package manager (rpm
> in Roger's case) that there is a full R installation available.

Really? Why should the user of the system have to accept a gratiuitous
install of about 100MB for just R itself, and perhaps 50MB is additional
unneeded packages without asking for them? How may the user know in
advance that extraneous dependencies have been added? Why then should that
user have to work out how to block extraneous installs of duplicate
software, and to do that before anyone has been told it will be required?
I do not think this is responsible packaging, it looks much more like
mission overreach.

Please tell me where to go to get includernw de-listed from
texlive-collection-mathscience and texlive-scheme-medium. Who maintains or
curates these lists, where may issues be raised?

Thanks for keeping with me.

As still partly responsible for the technical production of the R Journal,
and for maintenance of multiple R packages (themselves with many reverse
dependencies), I really do not need my workflows disrupted.

Roger

>
> But this is only necessary when disk space is scarce, what in the case
> of R should make a huge difference.
>
> Best
>
> Norbert
>
> --
> PREINING Norbert http://www.preining.info
> Accelia Inc. + JAIST + TeX Live + Debian Developer
> GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13
>

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: [phone]; e-mail: ***@nhh.no
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en

[Norbert Preining]

Hi Roger,

On Sun, 11 Nov 2018, Roger Bivand wrote:
> Really? Why should the user of the system have to accept a gratiuitous
> install of about 100MB for just R itself, and perhaps 50MB is additional
> unneeded packages without asking for them? How may the user know in advance

Well, that is the discrepancy that always will happen when you rescind
work to others, in this case the packagers.

You can set up your own slackware Linux, and install every single
package by hand. Then you can select *exactly* what you have installed.

But if you leave decisions to others, by using a distribution, the
distribution rules apply, even if you don't like them.

One of the rules most distributions have is that if you install a
package it is ready to use, and necessary dependencies are installed.

And for *most* users this is the *correct* way (I stand by this!). For
those who want something else, they need to invest time and energy, like
doing equivs packages, installing things only by hand, whatever.

> that extraneous dependencies have been added? Why then should that user have
> to work out how to block extraneous installs of duplicate software, and to

How should the packaging system GUESS what you have installed, and
whether it is in the right version etc etc.???

Sorry, you obviously have absoutely no idea about what the work of
distributors comprises, and instead of asking politely where to ask and
what to do, you are criticizing people who do an incredible hard work to
ship you a ready-made distribution.

I ask you a last question: Did you install a desktop environment, like
KDE or GNOME?

If yes, did you select *each*single*necessary* library and sub-program
that is necessary for it to run?

I guess you didn't. And there are for sure many packages installed (like
tracker, or zeitgeist, or whatever) that you maybe not know about and
never use.

And, did you complain?

It is only because you have your special setup that passes the
distribution packaging, and expect that distributors can conjure up some
magic that automatically agrees with your policy.

Please, get down to earth, and appreciate the work distributors, and
that includes us as TeX Live team, as well as all those doing packaging
work for Linux distributors, are doing.

> Please tell me where to go to get includernw de-listed from
> texlive-collection-mathscience and texlive-scheme-medium. Who maintains or
> curates these lists, where may issues be raised?

We told this to you already several times, it is Fedora. Use whatever
bug report mechanism is there. I don't use Fedora, I don't know, but
surely they have a bug tracking database.

> dependencies), I really do not need my workflows disrupted.

AGAIN!!! If you set up your stuff correctly, your workflow IS NOT
disrupted, because the R installation *you* have installed is found
first. If it is disrupted, then your setup is broken.

Thanks

Norbert

--
PREINING Norbert http://www.preining.info
Accelia Inc. + JAIST + TeX Live + Debian Developer
GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13

[Roger Bivand]

On Mon, 12 Nov 2018, Norbert Preining wrote:

> Hi Roger,
>
> On Sun, 11 Nov 2018, Roger Bivand wrote:
>> Really? Why should the user of the system have to accept a gratiuitous
>> install of about 100MB for just R itself, and perhaps 50MB is additional
>> unneeded packages without asking for them? How may the user know in advance
>
> Well, that is the discrepancy that always will happen when you rescind
> work to others, in this case the packagers.
>
> You can set up your own slackware Linux, and install every single
> package by hand. Then you can select *exactly* what you have installed.
>
> But if you leave decisions to others, by using a distribution, the
> distribution rules apply, even if you don't like them.

I accept that packages are rule makers and those installing packages are
rule takers in general.

I'm only questioning this particular case, as I'm convinced that, after an
R.core installation triggered by a dependency in includernw on R-knitr and
its R package dependency, trying to build the latex in R chunks in the Rnw
file will fail, because they in turn try to load other R packages that are
not available.

Yes, the includernw latex package can run its own example, but that is
about its limit unless the user already has an R installation.

>
> One of the rules most distributions have is that if you install a
> package it is ready to use, and necessary dependencies are installed.
>
> And for *most* users this is the *correct* way (I stand by this!). For
> those who want something else, they need to invest time and energy, like
> doing equivs packages, installing things only by hand, whatever.
>

I've bowed to your position as rule maker and mine as rule taker and used
PATH to avoid finding the spuriously installed R version.

Roger

>> that extraneous dependencies have been added? Why then should that user have
>> to work out how to block extraneous installs of duplicate software, and to
>
> How should the packaging system GUESS what you have installed, and
> whether it is in the right version etc etc.???
>
> Sorry, you obviously have absoutely no idea about what the work of
> distributors comprises, and instead of asking politely where to ask and
> what to do, you are criticizing people who do an incredible hard work to
> ship you a ready-made distribution.
>
> I ask you a last question: Did you install a desktop environment, like
> KDE or GNOME?
>
> If yes, did you select *each*single*necessary* library and sub-program
> that is necessary for it to run?
>
> I guess you didn't. And there are for sure many packages installed (like
> tracker, or zeitgeist, or whatever) that you maybe not know about and
> never use.
>
> And, did you complain?
>
> It is only because you have your special setup that passes the
> distribution packaging, and expect that distributors can conjure up some
> magic that automatically agrees with your policy.
>
> Please, get down to earth, and appreciate the work distributors, and
> that includes us as TeX Live team, as well as all those doing packaging
> work for Linux distributors, are doing.
>
>> Please tell me where to go to get includernw de-listed from
>> texlive-collection-mathscience and texlive-scheme-medium. Who maintains or
>> curates these lists, where may issues be raised?
>
> We told this to you already several times, it is Fedora. Use whatever
> bug report mechanism is there. I don't use Fedora, I don't know, but
> surely they have a bug tracking database.
>
>> dependencies), I really do not need my workflows disrupted.
>
> AGAIN!!! If you set up your stuff correctly, your workflow IS NOT
> disrupted, because the R installation *you* have installed is found
> first. If it is disrupted, then your setup is broken.
>
> Thanks
>
> Norbert
>
> --
> PREINING Norbert http://www.preining.info
> Accelia Inc. + JAIST + TeX Live + Debian Developer
> GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13
>

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: [phone]; e-mail: ***@nhh.no
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en

[Norbert Preining]

Hi Roger,

> I accept that packages are rule makers and those installing packages are
> rule takers in general.

It is a necessity. Consider the average linux distribution with about
20000 packages or more, where R is just a very small part of it.
One needs rules that govern the package usability.

> I'm only questioning this particular case, as I'm convinced that, after an
> R.core installation triggered by a dependency in includernw on R-knitr and
> its R package dependency, trying to build the latex in R chunks in the Rnw
> file will fail, because they in turn try to load other R packages that are
> not available.

This is then a packaging bug. If the maintainer wants to ensure full
functionality, the dependency on R-knitr needs to pull in either
directly or indirectly everything necessary for running at least a
minimal document (I don't say it needs to pull in everything that might
be possible at all under any conceivable situation, though!).

If even the most simple case of document cannot be compiled after
installation despite installing parts of R, this is a bug of the package
and should be reported to the maintainer.

> I've bowed to your position as rule maker and mine as rule taker and used
> PATH to avoid finding the spuriously installed R version.

That is not my rule. That has been the rule since the conceivment of
Unix, long before I was born.

Best

Norbert

--
PREINING Norbert http://www.preining.info
Accelia Inc. + JAIST + TeX Live + Debian Developer
GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13

[Roger Bivand]

Hi Norbert,

On Mon, 12 Nov 2018, Norbert Preining wrote:

> Hi Roger,
>
>> I accept that packages are rule makers and those installing packages are
>> rule takers in general.
>
> It is a necessity. Consider the average linux distribution with about
> 20000 packages or more, where R is just a very small part of it.
> One needs rules that govern the package usability.
>

OK. I also see that say texlive has internal dependencies between
components, like R. R has > 13000 contributed packages on CRAN, and almost
3000 more on Bioconductor, and we've been very concerned to prune reverse
dependency trees to limit installation of more than is strictly needed
even if some functionality is inhibited. We do this to permit nightly
cross-platform checking of all contributed packages hosted on CRAN.

>> I'm only questioning this particular case, as I'm convinced that, after an
>> R.core installation triggered by a dependency in includernw on R-knitr and
>> its R package dependency, trying to build the latex in R chunks in the Rnw
>> file will fail, because they in turn try to load other R packages that are
>> not available.
>
> This is then a packaging bug. If the maintainer wants to ensure full
> functionality, the dependency on R-knitr needs to pull in either
> directly or indirectly everything necessary for running at least a
> minimal document (I don't say it needs to pull in everything that might
> be possible at all under any conceivable situation, though!).
>
> If even the most simple case of document cannot be compiled after
> installation despite installing parts of R, this is a bug of the package
> and should be reported to the maintainer.

It passes this test (loading knitr), but almost any workflow will need
other sets of packages, depending on the R code in the Rnw document being
processed (which cannot be known before run time).

>
>> I've bowed to your position as rule maker and mine as rule taker and used
>> PATH to avoid finding the spuriously installed R version.
>
> That is not my rule. That has been the rule since the conceivment of
> Unix, long before I was born.

OK, my unix admin certification is from Sperry in 1985 (2MB memory, System
V). At the time, EMACS really did mean Eight Megabytes And Constant
Swapping.

Thanks for your patience,

Roger

>
> Best
>
> Norbert
>
> --
> PREINING Norbert http://www.preining.info
> Accelia Inc. + JAIST + TeX Live + Debian Developer
> GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13
>

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: [phone]; e-mail: ***@nhh.no
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en

[Norbert Preining]

Hi Roger,

On Mon, 12 Nov 2018, Roger Bivand wrote:
> OK. I also see that say texlive has internal dependencies between
> components, like R. R has > 13000 contributed packages on CRAN, and almost
> 3000 more on Bioconductor, and we've been very concerned to prune reverse
> dependency trees to limit installation of more than is strictly needed even

CRAN is in a very good situation due to its very strict rules concerning
upload format. CTAN is older and has historically very loose rules,
which CTAN does not want to change now. That means, we have practically
no way of testing.

Alone converting CTAN packages to standard TDS standard as used in TeX
Live (and MikTeX and any other TeX distribution I know) is a crazy task
(see our ctan2tds script), which requires hand-work for each and every
package.

(I sometimes *DREAM* of the level of consistency CRAN has when seeing what
is shipped on CTAN and we need to incorporate)

> if some functionality is inhibited. We do this to permit nightly
> cross-platform checking of all contributed packages hosted on CRAN.

That is something impossible on CTAN level. I am working on doing
something similar on TeX Live level, but this, too, is a very bad task
due to the variety of engines (latex, pdflatex, platex, lualatex,
xetex, uptex, ......) that might be necessary to actually run a test
fileusing a certain package.

> It passes this test (loading knitr), but almost any workflow will need other
> sets of packages, depending on the R code in the Rnw document being
> processed (which cannot be known before run time).

That is fine. Maintainers cannot and should not cater to any possible
usage at all. I (putting my Debian maintainer hat on) often tell people
that with only texlive-base installed, they can compile simple TeX
documents, but nothing more. But this is it. If they want more, the need
to install texlive-*-recommended etc etc.

Best

Norbert

--
PREINING Norbert http://www.preining.info
Accelia Inc. + JAIST + TeX Live + Debian Developer
GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13

[Roger Bivand]

On Sun, 11 Nov 2018, Andreas Storvik Strauman wrote:

> As I have come to understand it, the problem is that Fedora
> installs R together with texlive-collection-mathscience because of
> includernw.sty, due to it's dependency on R and R-knitr.
>
> However, Roger does not want to use includernw, and has R installed
> (in a non-conventional location on his computer). This leaves Roger
> with two R installations and a .sty he doesn't use.
>
> Am I correct, Roger?

Yes, correct. And this is simply from upgrading texlive, so nothing that I
have opted into (apart from installing texlive-scheme-medium).

Roger

>
> ________________________________
> From: Norbert Preining <***@logic.at>
> Sent: Sunday, November 11, 2018 3:14:29 PM
> To: Roger Bivand
> Cc: Andreas Storvik Strauman; tex-***@tug.org
> Subject: Re: [tex-live] inclusion of texlive-includernw
>
> Hi Roger,
>
> I have read now both your answers, but still tap in the dark with what
> you want to achieve and what is your actual problem. Your descriptions
> are a complete mixture of nomenclature.
>
> On Sun, 11 Nov 2018, Roger Bivand wrote:
>> No, not every. The R package manager (so the R code in this latex package)
>> advises limiting the number of dependencies (Depends, Imports) as far as
>
> I don't know what you are talking about, honestly.
>
> The latex package you are refering to does *NOT* contain any package
> manager facilities. I calls R on some code which calls
> "library('knitr')".
>
> You are mixing completely different things here: depends, suggests etc
> are entities of the fedora packaging, and have nothing to do with the
> latex package.
>
> Upstream (other of includenrw) has no influence whatsoever what fedora
> packager are doing.
>
>> So "correct" has many flavours. I will try to trace the Fedora packager, and
>> follow up with the author of includernw, who certainly did not ask for the
>> latex package to be rolled out at least to all texlive-scheme-medium users
>> on Fedora.
>
> Again wrong. includenrw is included in TeX Live. And the Fedora
> maintainer has packaged TeX Live and thus also includenrw.
> There is nothing wrong, and if at all, then it is in your handling of
> how you install R packages and manage them.
>
> It is *you* who has to decide whether you want to use the fedora
> packaged versions of R pakcages, or locally isntalled ones.
>
> BTW, I know what I am talking about: I am TeX Live developer, at the
> same time maintainer of TeX Live in Debian, and do quite a lot of work
> with R. So trust me, it is in *your* obligation to keep a workable
> environment.
>
> If you start installing local R packages you take the control intio your
> hands, but also the responsability to have all the necessary tools etc
> installed.
>
> Don't blame either the author of includenrw or the fedora maintainer for
> problems.
>
> Best
>
> Norbert
>
> --
> PREINING Norbert http://www.preining.info
> [http://www.preining.info/front-page.png]<http://www.preining.info/>
>
> Norbert Preining<http://www.preining.info/>
> www.preining.info
> latest galleries: (c) 2012-2018 Norbert Preining
>
>
>
> Accelia Inc. + JAIST + TeX Live + Debian Developer
> GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13
>

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: [phone]; e-mail: ***@nhh.no
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en

[Pablo Alfonso González Luengo]

The best, in many cases, is not to depend on the version installed by the
package manager of the distribution. In my case equal use fedora 29
(updated 28 to version 29). This works for me:
1. Remove all packages related to texlive installed by dnf
2. Execute the following lines
$ sudo dnf copr enable fatka/texlive-dummy
$ sudo dnf -y install texlive-dummy
If it fails for some reason, manually download and install
"texlive-dummy-9999-4.noarch.rpm"
3. Download and install the official version of TexLive 2018 from
http://mirror.ctan.org/systems/texlive/tlnet/install-tl-unx.tar.gz
4. Prepare a coffee (or several) ...wait...wait...done.
This has always worked for me in relation to TexLive