Pandoc: Raw vs parsed metadata fields

[dato]

Hello!

I have the following use-case of _header-include_:

title: Pandoc bug?
header-includes: \pgfpagesuselayout{2 on 1}[a4paper]

---

Test.

When generating a PDF, the string [a4paper] gets converted to the invalid LaTeX sequence {[}a4paper{]}:

% pandoc -s -w latex <foo.md | grep uselayout
\pgfpagesuselayout{2 on 1}{[}a4paper{]}

Am I doing something wrong?

I also tried the following variations, to no avail:

title: Pandoc bug?
header-includes: |
  \usepackage{pgfpages}
  \pgfpagesuselayout{2 on 1}[a4paper]

---

Test.

And:

title: Pandoc bug?
header-includes:
  - \usepackage{pgfpages}
  - \pgfpagesuselayout{2 on 1}[a4paper]

---

Test.

May thanks in advance! My pandoc version is 1.13.2.

[dato]

(Note that this doesn't happen when using --include-in-header from the command line.)

[jgm]

Here's a clue:

% pandoc -t native   # [edit: fixed command line]
 \pgfpagesuselayout{2 on 1}[a4paper]
[Para [RawInline (Format "latex") "\\pgfpagesuselayout{2 on 1}",Str "[a4paper]"]]

The basic problem is that pandoc has no way of knowing that the command \pgfpagesuselayout consumes an optional argument after its first argument. It guesses, incorrectly, that [a4paper] is regular text, not part of a LaTeX command. And then, when the regular text brackets are written back out to LaTeX, they're escaped.

[jgm]

When you use --include-in-header from the command line, the line is included verbatim, not parsed as Markdown. Metadata fields are parsed as Markdown.

[lierdakil]

Makes sense. As I see it, there are two ways to handle this: either modify
Markdown parser to understand this type of LaTeX commands, or invent some
syntax to allow inclusion of verbatim strings in metadata. Former would be
better generally, but may be a bit fragile, I think. Latter sounds like a
hack, but may be useful in some cases. Are there any reservations for any
of these options?

2015-05-07 19:16 GMT+03:00 John MacFarlane [email protected]:

When you use --include-in-header from the command line, the line is
included verbatim, not parsed as Markdown. Metadata fields are parsed as
Markdown.

—
Reply to this email directly or view it on GitHub
https://github.com/jgm/pandoc/issues/2139#issuecomment-99925776.

[jgm]

+++ Nikolay Yakimov [May 07 15 09:25 ]:

Makes sense. As I see it, there are two ways to handle this: either
modify
Markdown parser to understand this type of LaTeX commands, or invent
some
syntax to allow inclusion of verbatim strings in metadata. Former would
be
better generally, but may be a bit fragile, I think. Latter sounds like
a
hack, but may be useful in some cases. Are there any reservations for
any
of these options?

I think include-in-header really only makes sense as verbatim, so one option would be to special-case that metadata field, though that seems a bit unprincipled.

[dato]

Wow, I hadn’t realized that the metadata block, including _header-includes_, was parsed as Markdown.

From where I stand, I agree that _header-includes_ only makes sense as verbatim; but I don't know if there'd be other bizarre use cases out there.

All in all, I'd appreciate if I could use this, whichever way, without having to resort to an external header file.

Many thanks in advance!

[edited: s/include-in-header/header-includes/]

[lierdakil]

I hacked together a proof-of-concept code for inclusion of "raw strings" in metadata. Current work can be found on lierdakil/pandoc:rawstring-metadata branch (diff)

This hacks on YAML object syntax.

Example:

pandoc -t native -s 
---
header-includes:
  rawstring: \pgfpagesuselayout{2 on 1}[a4paper]
...

Test
^D
Pandoc (Meta {unMeta = fromList [("header-includes",MetaString "\\pgfpagesuselayout{2 on 1}[a4paper]")]})
[Para [Str "Test"]]

I'd like some comments on the viability of this idea.

Another option is to add new flag to ReaderOptions (e.g. optRawMetaStrings :: Bool), and implicitly set it for select few metadata fields. Maybe we could also add an explicit CLI argument (e.g. --raw-metadata-strings) to set this option for all metadata, but I'm not convinced that would be useful.

[jgm]

+++ Nikolay Yakimov [May 14 15 20:44 ]:

I hacked together a proof-of-concept code for inclusion of "raw strings" in metadata. Current work can be found on lierdakil/pandoc:rawstring-metadata branch (diff)

This hacks on YAML object syntax.

Example:

pandoc -t native -s
---
header-includes:
 rawstring: \pgfpagesuselayout{2 on 1}[a4paper]
...

Test
^D
Pandoc (Meta {unMeta = fromList [("header-includes",MetaString "\\pgfpagesuselayout{2 on 1}[a4paper]")]})
[Para [Str "Test"]]

I'd like some comments on the viability of this idea.

I think that ultimately we'll want this to be customizable
per output format. Your proposal gives you a document that
can only be rendered in one format, which is sort of against
the spirit of pandoc.

Your proposal does have the advantage of not needing
changes to pandoc-types (e.g. addition of a new metadata type that
defines raw string output for each format). But I'm not sure
we should go for a half-solution.

I suppose we could implement something now that could be naturally
extended in the direction I suggested. For example:

header-includes:
  (*):  raw stuff

((*) means: raw content in any output format. Yes, this actually
works as a field name!) Then in the future we could extend:

header-includes:
  (latex):  '\raw{latex}'
  (html):   '<raw>html</raw>'
  (*):      'raw fallback'

Another option is to add new flag to ReaderOptions (e.g. optRawMetaStrings :: Bool), and implicitly set it for select few metadata fields. Maybe we could also add an explicit CLI argument (e.g. --raw-metadata-strings) to set this option for all metadata, but I'm not convinced that would be useful.

This seems less useful to me.

[lierdakil]

The more I think about your idea, the more I like it. And parenthesis syntax looks reasonable enough, I definitely prefer it to initially-proposed leading-underscore syntax.

Now as for implementation, I think it's entirely possible to do without changes to pandoc-types, actually, since we already have MetaMap.

It should be possible (and not that horrible) to hack templating engine, so that if a variable resolves to Object, it would try to select from it based on output format name, if that fails on wildcard, and if that fails, default to "true" (as it does now, if I'm not mistaken). Adding a new constructor to MetaValue wouldn't add anything new to this general algorithm, since it's converted to JSON anyway.

It's not entirely clear if we should do something similar when passing metadata to filters, or just pass these "format-dependent raw string" objects in their entirety.

Thoughts?

[lierdakil]

So, I've played around with this idea for a little bit. See ec17fff for proof-of-concept implementation of above proposal. Code does need some polish, nevermind docs and tests, but general concept should be possible to grasp.

I struggle a bit with HTML-based output though, since only HTML writer actually uses template engine, so we'll have to pass writer type to writeHtml if we want to parametrize for other HTML-based formats, like EPUB (it's possible to keep API relatively intact by introducing new function, of course, but I'd like to have some feedback before getting on with it)

Also, I'm not sure about my implementation of WriterType. On one hand, it should be possible to delegate these declarations to writers (using classes and probably Data.Typeable), on the other hand, it makes this either considerably more verbose (and less type-safe), or considerably less flexible (i.e. at the moment, it's relatively simple to use multiple prioritized keys for single writer -- one example where it could be useful is, again, HTML-based output, not sure on how to implement similar concept with classes)

[bpj]

This can be worked around with a simple filter which overloads flagged code/codeblock elements.
https://gist.github.com/bpj/e6e53cbe679d3ec77e25#file-pandoc-code2raw-py

[yihui]

I have been bitten by this issue a couple of times. Today I tried to set mainfontoptions in the YAML metadata like this:

---
mainfont: Alegreya
mainfontoptions: "UprightFeatures={SmallCapsFont=AlegreyaSC-Regular}"
---

It didn't work because Pandoc escaped the curly braces:

\setmainfont[UprightFeatures=\{SmallCapsFont=AlegreyaSC-Regular\}]{Alegreya}

I know this has been reported in #2565 -- just one more vote, hoping raw YAML fields could be possible some day.

[nichtich]

Why not make use of YAML tags? By now these tags are are ignored, so

---
foo: !!whatever "*bar*"
...

is parsed equivalent to

---
foo: *bar*
...

as

Pandoc (Meta {unMeta = fromList [("foo",MetaInlines [Emph [Str "bar"]])]})

Metadata values passed on command line are not parsed in Markdown but end up as MetaString (or MetaBool). The former should also settable in YAML like this:

---
foo: !!MetaString "*bar*"
...

as

Pandoc (Meta {unMeta = fromList [("foo",MetaString "*bar*")]})

This does not solve the full issue but at least gives a workaround. Last but not least disabling of Markdown parsing for selected YAML fields makes sense in other use cases too.

P.S: It would be consistent to also support MetaBool but actually not needed much.

---
foo: !!MetaBool "true"
...

should be parsed like

---
foo: true
...

[jgm]

That's a really nice idea.

+++ Jakob Voß [Sep 23 16 00:26 ]:

Why not make use of YAML tags? By now these tags are are ignored, so

---

foo: !!whatever "_bar_"
...

is parsed equivalent to

---

foo: _bar_
...

as
Pandoc (Meta {unMeta = fromList [("foo",MetaInlines [Emph [Str "bar"]])]})

Metadata values passed on command line are not parsed in Markdown but
end up as MetaString (or MetaBool). The former should also settable in
YAML like this:

---

foo: !!MetaString "_bar_"
...

as
Pandoc (Meta {unMeta = fromList [("foo",MetaString "_bar_")]})

This does not solve the full issue but at least gives a workaround.
Last but not least disabling of Markdown parsing for selected YAML
fields makes sense in other use cases too.

—
You are receiving this because you commented.
Reply to this email directly, [1]view it on GitHub, or [2]mute the
thread.

References

1. https://github.com/jgm/pandoc/issues/2139#issuecomment-249121649
2. https://github.com/notifications/unsubscribe-auth/AAAL5DJOcw8Htvi5f0IlDSxAyuW1eCq2ks5qs38ogaJpZM4ESjxf

[bpj]

Just two observations:

1. It may be better to use local tags !MetaMap etc. so that the YAML has
   a chance of parsing 'out of context' without having a formal tag definition
   included.
2. There should also be a tag !Data or the like to indicate that the
   whole tree below it should be made available to filters as a plain JSON
   structure without any interpretation as Markdown. That would make filter
   configuration much easier.

Den 23 sep 2016 10:13 skrev "John MacFarlane" [email protected]:

That's a really nice idea.

+++ Jakob Voß [Sep 23 16 00:26 ]:

Why not make use of YAML tags? By now these tags are are ignored, so

---

foo: !!whatever "_bar_"
...

is parsed equivalent to

---

foo: _bar_
...

as
Pandoc (Meta {unMeta = fromList [("foo",MetaInlines [Emph [Str "bar"]])]})

Metadata values passed on command line are not parsed in Markdown but
end up as MetaString (or MetaBool). The former should also settable in
YAML like this:

---

foo: !!MetaString "_bar_"
...

as
Pandoc (Meta {unMeta = fromList [("foo",MetaString "_bar_")]})

This does not solve the full issue but at least gives a workaround.
Last but not least disabling of Markdown parsing for selected YAML
fields makes sense in other use cases too.

—
You are receiving this because you commented.
Reply to this email directly, [1]view it on GitHub, or [2]mute the
thread.

References

1. https://github.com/jgm/pandoc/issues/2139#issuecomment-249121649
2. https://github.com/notifications/unsubscribe-auth/
   AAAL5DJOcw8Htvi5f0IlDSxAyuW1eCq2ks5qs38ogaJpZM4ESjxf

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/jgm/pandoc/issues/2139#issuecomment-249130371, or mute
the thread
https://github.com/notifications/unsubscribe-auth/ABG3Uy5bCI_lF-zAaKmXCB7GSdUsIl1Iks5qs4pFgaJpZM4ESjxf
.

[lierdakil]

A couple observations of my own:

1. Data.Yaml doesn't support yaml tags, unless using experimental and
   unstable (as of yaml-0.8.18.7) Data.Yaml.Parser. And even ignoring that
   it's pretty awkward to use.
2. Data.Libyaml does support yaml tags, but it's a low-level streaming
   parser based on pipes, so it's kinda complicated.

So, at the moment, there seems to be no concise way of getting at yaml tags
from Haskell. So while the idea of using yaml tags is good on paper,
implementation... uh... won't be pretty, that's for sure.

2016-09-23 22:43 GMT+03:00 Benct Philip Jonsson [email protected]:

Just two observations:

1. It may be better to use local tags !MetaMap etc. so that the YAML has
   a chance of parsing 'out of context' without having a formal tag definition
   included.
2. There should also be a tag !Data or the like to indicate that the
   whole tree below it should be made available to filters as a plain JSON
   structure without any interpretation as Markdown. That would make filter
   configuration much easier.

Den 23 sep 2016 10:13 skrev "John MacFarlane" [email protected]:

That's a really nice idea.

+++ Jakob Voß [Sep 23 16 00:26 ]:

Why not make use of YAML tags? By now these tags are are ignored, so

---

foo: !!whatever "_bar_"
...

is parsed equivalent to

---

foo: _bar_
...

as
Pandoc (Meta {unMeta = fromList [("foo",MetaInlines [Emph [Str
"bar"]])]})

Metadata values passed on command line are not parsed in Markdown but
end up as MetaString (or MetaBool). The former should also settable in
YAML like this:

---

foo: !!MetaString "_bar_"
...

as
Pandoc (Meta {unMeta = fromList [("foo",MetaString "_bar_")]})

This does not solve the full issue but at least gives a workaround.
Last but not least disabling of Markdown parsing for selected YAML
fields makes sense in other use cases too.

—
You are receiving this because you commented.
Reply to this email directly, [1]view it on GitHub, or [2]mute the
thread.

References

1. https://github.com/jgm/pandoc/issues/2139#issuecomment-249121649
2. https://github.com/notifications/unsubscribe-auth/
   AAAL5DJOcw8Htvi5f0IlDSxAyuW1eCq2ks5qs38ogaJpZM4ESjxf

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/jgm/pandoc/issues/2139#issuecomment-249130371, or
mute
the thread
zAaKmXCB7GSdUsIl1Iks5qs4pFgaJpZM4ESjxf>
.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/jgm/pandoc/issues/2139#issuecomment-249284852, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AG8EZi7-LlCSnVrSShRor3PEF94wjzO4ks5qtCvfgaJpZM4ESjxf
.

[jgm]

OK, I've lost enthusiasm for the Yaml tags idea.
I still like the idea I expressed above, of using objects with parenthesized fields for raw content.
Thus, for example,

    header-includes:
      - (latex):  '\raw{latex}'
        (html):   '<raw>html</raw>'
        (*):      'raw fallback'

or

mainfontoptions:
  (latex): "UprightFeatures={SmallCapsFont=AlegreyaSC-Regular}"

This would require the following changes, I think:

• The code in the Markdown reader for parsing YAML metadata would parse as a raw MetaString any field name surrounded by parentheses.
• The template expansion code would, when rendering an object, check to see if it has a field named (FORMAT), where FORMAT is the current output format. If so, render that content. If not, check for (*), and render it if found. Otherwise treat like other objects.

This could be done without touching pandoc-types, I believe. More ambitiously we could change MetaString to a map from formats to raw strings.

[jgm]

(_) or () would be alternatives to (*) for the fallback.

[jgm]

Another option would be to modify the Markdown reader so it parses

mainfontoptions:
  (latex): "UprightFeatures={SmallCapsFont=AlegreyaSC-Regular}"

as metafontoptions with a RawBlock (Format "latex") "the string".
Then no modifications to the template engine would be needed.

[jgm]

To put yet another option on the table (again involving on the Markdown reader):

mainfontoptions[latex]: "UprightFeatures={SmallCapsFont=AlegreyaSC-Regular}"

This would just require checking the key; if it ends with [format], then we parse it as RawInline (Format format) (or perhaps RawBlock? It might not matter in this context.)

This would do the right thing in most cases, e.g. the raw latex would be omitted automatically in HTML output. And it would be easy to implement and easy to remember.

[nichtich]

I prefer the latter format with no addition nesting but I can't say why ;-) mainfontoptions@latex may be another alternative. What woul be the metadata for the following document for different output formats?

foo[latex]: a
foo[html]: b
foo[json]: c
foo[native]: d
foo[notexist]: e
foo[]: f
foo: g

I'd guess that it's

• fromList [("foo", RawInline (Format "latex") "a")] for latex
• fromList [("foo", RawInline (Format "html") "b")] for html
• fromList [("foo",MetaInlines [Str "g"])] for all other formats, including json and native.

In any way the AST only contains a single metadata field foo, right?

[jgm]

I'd guess that it's
* fromList [("foo", RawInline (Format "latex") "a")] for latex
* fromList [("foo", RawInline (Format "html") "b")] for html
* fromList [("foo",MetaInlines [Str "g"])] for all other formats,
including json and native.

No, the parser doesn't know what the output format is going
to be, so it would have to be something like

 fromList [("foo", [MetaInlines [RawInline (Format "latex") "a",
                                 RawInline (Format "html") "b"],
                     MetaString "f"])]

There are obviously some complexities here to consider...

[bpj]

@jgm wrote:

(_) or () would be alternatives to (*) for the fallback.

Are you aware that the asterisk is a metacharacter in YAML? All keys containing it would have to be quoted. The underscore doesn't have that issue.

[jgm]

I think that with 2b34337a9cf8b025914e8219498b4c0258772be0 we now have a decent solution to this problem.

title: Pandoc bug?
header-includes: `\pgfpagesuselayout{2 on 1}[a4paper]`{=latex}
---

    header-includes:
      - `\raw{latex}`{=latex}
        `<raw>html</raw>`{=html}

[jgm]

Please comment here if you think that we should still support a syntax like

    header-includes:
      - (latex):  '\raw{latex}'
        (html):   '<raw>html</raw>'
        (*):      'raw fallback'

now that similar things can be achieved with the raw_attribute.

[yihui]

I think at least my problem will be solved by the new raw attribute. Thanks!

[jgm]

I'm going to close this unless anyone really wants to bring it back.

[pauljohn32]

How to use this? My short run goal is to make this work in the YAML header

logoleft: "/home/pauljohn/R/x86_64-pc-linux-gnu-library/3.4/crmda/theme/jayhawk.pdf"

With pandoc 1.19, I get "_" in the tex file.

My long run goal is to replace the explicit path with the result of an R chunk which retrieves a file path. This does work to retrieve the path, I can use it in export to HTML, but not PDF because "_" is inserted:

logoleft: "`r system.file('theme/jayhawk.pdf', package = 'crmda')`"

That works if package is installed in /usr/local/share, for example. However, if the R package is in user directory, where there is "_", its a total fail.

[jgm]

logoleft: "`r system.file('theme/jayhawk.pdf', package = 'crmda')`{=html}"

But this only works with the dev version of pandoc, not with 1.19.

[Merovex]

I ran into this problem recently. I solved it by simply adding my metadata.yml file as the first of the array of input files.

pandoc -t beamer -o build/presentation.pdf metadata.yml slides.md

[huyhoang8398]

%change color fot the institute
\setbeamercolor{institute in head/foot}{fg=black}
\makeatletter
%change the footline template to include frame numbers
\defbeamertemplate*{footline}{myminiframes theme}
  {%
    \begin{beamercolorbox}[colsep=1.5pt]{upper separation line foot}
    \end{beamercolorbox}
    \begin{beamercolorbox}[ht=2.5ex,dp=1.125ex,%
      leftskip=.3cm,rightskip=.3cm plus1fil]{author in head/foot}%
      \leavevmode{\usebeamerfont{author in head/foot}\insertshortauthor}%
      \hfill%
      {\usebeamerfont{institute in head/foot}\usebeamercolor[fg]{institute in head/foot}\insertshortinstitute}%
    \end{beamercolorbox}%
    \begin{beamercolorbox}[ht=2.5ex,dp=1.125ex,%
      leftskip=.3cm,rightskip=.3cm plus1fil]{title in head/foot}%
      {\usebeamerfont{title in head/foot}\insertshorttitle\hfill \insertframenumber/\inserttotalframenumber}%<-here
    \end{beamercolorbox}%
    \begin{beamercolorbox}[colsep=1.5pt]{lower separation line foot}
    \end{beamercolorbox}
  }
\makeatother

%change look of sections in ToC
\defbeamertemplate*{section in toc}{mysections in toc}
{\leavevmode ---\,\inserttocsection\par}

%change look of subsections in ToC
\defbeamertemplate*{subsection in toc}{mysections in toc}
{\leavevmode\leftskip=2.5em --\,\inserttocsubsection\par}

I want to customize Ilmenau beamer theme ( add the slide number into the footer ). I tried to add Raw latex with -header-includes but it said that they cannot parse YAML metadata.

[pauljohn32]

I do not use llmenau beamer theme, but I can show you a preamble code section that you can adjust to suit your needs. In the YAML header, include a code block like so:

output:
  beamer_presentation:
    theme: "kucrmda"
    keep_tex: false
    latex_engine: pdflatex
    citation_package: natbib
    includes:
      in_header: preamble-listings.tex

This means that in your current directory, a file "beamerthemekucrmda.sty" must be found, and we also
have a preamble for general purpose latex code you may not need under includes called "preamble-listings.tex".

In the "beamerthemekucrmda.sty" file, here is example. Look down to the last line.

\mode<presentation>
\definecolor{kublue}{RGB}{0,81,186}
\usefonttheme{professionalfonts}
\useoutertheme{infolines}
\useinnertheme{rounded}
%disable rounded for alert and example boxes%
\setbeamertemplate{blocks}[default]
\usecolortheme{whale}
\usecolortheme{orchid}
\setbeamercolor{structure}{bg=kublue,fg=kublue!90!black}
\setbeamercolor{frametitle}{bg=kublue}
\setbeamercolor{section in toc}{fg=kublue!40!black}

\setbeamertemplate{frametitle continuation}[from second]
\renewcommand\insertcontinuationtext{...}
\beamertemplatenavigationsymbolsempty

\expandafter\def\expandafter\insertshorttitle\expandafter{%
 \insertshorttitle\hfill\insertframenumber\,/\,\inserttotalframenumber}

\mode<all>

Do you see in the very last provision, we have inserted the "frame number, total frame number" piece at the end. I suspect that if you just put in that part, which goes with "short title", you will have good result. Only difficult part is reading your theme to figure out what name they use to refer to the part where you want to insert the number. I predict you will succeed!

[huyhoang8398]

Thank you so much @pauljohn32, it worked like a charm.
I also found another solution for those who want to use beamer theme with miniframes on the top and also the footer with the frame number,
You can use CambrideUS theme, and in you YAML file, set an variable

theme:
- CambridgeUS 
colortheme:
- default 
innertheme:
- rounded 
outertheme:
- miniframes 

So with the outertheme: miniframes, you will get the miniframe and also the default footer of cambrideUS and you dont have to use an sty file because the CambrideUS is a theme Without Navigation Bar.
So the result will be like this pic

[pauljohn32]

Just for the record, there is no magic in this. Look at the file "beamerthemeCambridgeUS.sty", you see the outertheme is "infolines". Then look at "beamerouterthemeinfolines.sty" where you find the design for the top and bottom banners.

\defbeamertemplate*{footline}{infolines theme}
{%
  \leavevmode%
  \hbox{%
  \begin{beamercolorbox}[wd=.333333\paperwidth,ht=2.25ex,dp=1ex,center]{author in head/foot}%
    \usebeamerfont{author in head/foot}\insertshortauthor\expandafter\ifblank\expandafter{\beamer@shortinstitute}{}{~~(\insertshortinstitute)}
  \end{beamercolorbox}%
  \begin{beamercolorbox}[wd=.333333\paperwidth,ht=2.25ex,dp=1ex,center]{title in head/foot}%
    \usebeamerfont{title in head/foot}\insertshorttitle
  \end{beamercolorbox}%
  \begin{beamercolorbox}[wd=.333333\paperwidth,ht=2.25ex,dp=1ex,right]{date in head/foot}%
    \usebeamerfont{date in head/foot}\insertshortdate{}\hspace*{2em}
    \usebeamercolor[fg]{page number in head/foot}\usebeamerfont{page number in head/foot}\usebeamertemplate{page number in head/foot}\hspace*{2ex} 
  \end{beamercolorbox}}%
  \vskip0pt%
}

The Beamer themes are surprisingly simply, easy to adjust to your purpose. If you liked the other theme you were using, adjust it...

[alerque]

Although not reflected in the discussion above, apparently some form of this was later implemented. See #5673.

[lierdakil]

@alerque, @jgm mentioned raw attributes are implemented in https://github.com/jgm/pandoc/commit/2b34337a9cf8b025914e8219498b4c0258772be0 above. So not sure where you got that "not reflected in the discussion above" thing.