bugGNU roff - Bugs: bug #58796, preconv: want option to write...

 
 

bug #58796: preconv: want option to write traditional [g|t]roff special characters where possible

Submitter:  G. Branden Robinson <gbranden>
Submitted:  Tue 21 Jul 2020 11:27:56 AM UTC
   
 
Category:  Preprocessor preconv Severity:  1 - Wish
Item Group:  Feature change Status:  Need Info
Privacy:  Public Assigned to:  gbranden
Open/Closed:  Open Planned Release:  None
* Mandatory Fields

Add a New Comment Rich Markup
   

Jump to the original submission

Thu 06 Aug 2020 03:54:27 PM UTC, comment #6: 

Werner posted to the mailing list to quickly set me straight:

> Pfft.  If you search for 'normalization' in the groff info manual, you
> will find



  For simplicity, all Unicode characters that are composites must be
  decomposed maximally (this is normalization form D in the Unicode
  standard);


> :-)


Thanks!  I'm glad to be wrong, as NFD is my favorite anyway.

Please excuse my ignorance--I haven't gotten around to messing up that part of our Texinfo manual yet.  ;-)

G. Branden Robinson <gbranden>
Group administrator
Thu 06 Aug 2020 02:35:20 PM UTC, comment #5: 

Just a couple of miscellaneous data points here.

1. "I neither think that groff is really C++, it is more like C with some aspects of classes"

My understanding from perusal of the sources over the past few years and a hazy recollection of Stroustrup 2nd edition is that groff is written in a limited subset of C++ as it existed in 1990 or so.  Maybe a little past the 1st edition of the C++ book, but not a lot.  Templates are not used anywhere I've seen.  None of the funny cast operators <static_cast>, <const_cast>, <reinterpret_cast>, <dynamic_cast> appear to be present.

There are still comments in the source code that refer to workarounds for CFRONT.

2. Something no one ever seems to mention when talking about supporting Unicode natively is which Normalization Form(s) we should support.

G. Branden Robinson <gbranden>
Group administrator
Wed 05 Aug 2020 01:50:29 PM UTC, comment #4: 

Regarding note #3 (it's all a bit tangential to this ticket):

I meant requiring iconv(3) at compile time, not iconv(1) at run-time.  The former would be horrible, the latter almost harmless.

I neither think that groff is really C++, it is more like C with some aspects of classes, nor am i aware of features in C++ to handle Unicode.  When you handle Unicode in C++, you just use C library features.  And Unicode handling in Perl and C is totally different, so much so that i can hardly think of any commonalities, so talking about Perl is really pointless here.

C features for handling Unicode are so bad that even for a new program, i would seriously consider making it ASCII only even today rather than using them.  When Kristaps started mandoc ten years ago, he made exactly that decision, and i'm very grateful to him for that, is was a very wise decision.  Even though internally, it uses practically no C-library Unicode-handling features and no other Unicode-handling library, it has practically perfect and in particular extremely robust and simple Unicode support.  I very much doubt that converting an existing program like groff would be a good idea, even if you had time to waste for a purely make-work project, in particular since groff already has a way for handling Unicode that works reasonably well and is simpler and more robust than anything you could do with the native C library features.

Of course, all this turns out to argue slightly in favour of Branden's idea, but i still think this ticket isn't high priority.

Ingo Schwarze <schwarze>
Group Member
Thu 30 Jul 2020 05:11:08 PM UTC, comment #3: 

Thanks for the comments, Ingo.  I understand and support the Unix philosophy, but I disagree with some of your underlying assumptions.

If you developed a brand-new tool to do some text-processing task, something designed to be used in pipelines with other tools, you could choose to specify that:
  a) the input character set of your tool be a Unicode encoding, or
  b) the tool only take some subset of Unicode as input, and require another tool to pipe in translations for the rest of Unicode, using a syntax invented specifically for these tools and not standardized anywhere else.

If you chose (b) on the grounds "pipelines are more Unixy," this would not be a popular choice.  Requiring helper applications to understand modern character sets is not inherently "the Unix way."  It's a stopgap used for historical applications whose cores do not (yet) speak Unicode.

Groff is a historical application.  It will always support \['e] because it must always be able to process historical documents that used such character representations.  But \['e] should in no way be considered the canonical way to represent the Unicode character LATIN SMALL LETTER E WITH ACUTE.  Unicode gives us the canonical representation.  \['e] and \[u00E9] are merely additional, roff-specific ways to represent this character.

The "roff-specific" part is important: the entire Unix philosophy of pipelines requires that all I/O be in as general a form as possible to be able to interact with as wide a range of other programs as possible.  groff and preconv, by contrast, communicate in a secret code that no other tool uses.  That's not the Unix way; that's a band-aid to cover up something that Werner identified as one of the four major areas of groff that needed to be updated back in 2013.  The need has not lessened in the intervening years.

That groff is a historical package does not absolve it from modern best practices in software design.  Looking to the long term, this is what we should be striving for.  preconv is a very useful bridge in the meantime; I believe you that the task of converting historical C++ code to natively handle UTF-8 input is big and messy.*  Nonetheless it should be considered groff's ultimate goal.

 * I'm currently going through a similar process--on a much smaller scale--with some Perl code.  And Perl actually handles a lot of the logic automatically that a C program would have to manually implement.  I don't know what C++'s facilities are like, but I do know that no matter how good the language's design, you'll run into stupid problems that will derail you for a few hours.

comment #2:

> I would hate it if groff would start requiring iconv.


It's far better to leverage existing code that does what you need than to re-implement the same logic in your own code.  The principle "solve one task only, but solve it well" ought to free the groff package from implementing its own conversions between character encodings and let it instead focus on its primary task.

Anyway, if groff handled Unicode I/O natively (and thus also ASCII, a subset thereof), I wouldn't expect iconv to become an installation requirement; it would be a run-time requirement for those users who need to feed in documents in other character encodings.

> it's much better to encode all non-ASCII characters and not force users to adopt an obsolete locale.


Good points here; I agree.  I fell into the trap of looking at the encoding groff currently natively handles, and not at the big picture.

Dave <barx>
Group Member
Sat 25 Jul 2020 08:41:12 PM UTC, comment #2: 

Hi Dave,

> a bit of a hack


Not so much, actually.  Making good use of pipes is among the design principles of the whole roff ecosystem, to harmonize with the overall UNIX design philosophy that every tool should solve one task only, but solve it well and in a way that facilitates combination with the other tools.  In this sense, groff is actually more UNIXy than mandoc, which does integrate preconv.

> wrapper for iconv


I would hate it if groff would start requiring iconv.  I consider it an important asset that so far, it does not.

> the language has standard libraries to handle UTF-8


Yes, indeed the C language contains a vast array of C library functions to deal with wide characters and with multibyte characters.  But the design of these C libary facilities is atrocious, and using something else which is non-standard would even be worse.  Either way, rewriting a program to natively support wide characters is usually an extremely tedious, extremely intrusive, very time-consuming and highly error-prone task.  Even when done as designed, it adds horrible complication to the code and makes the code much more fragile.  For samll programs, ways exist to cheat one's way around these notorious downsides, see my presentation at EuroBSDCon in Beograd a few years ago.  But i doubt something like that could be pulled off for a program as large as groff, at least not easily.

> not sure why preconv need emit things like \['e] or \[u00E9] at all


Because single-byte 8-bit locales have been obsolete for many years and some operating systems don't even support them any longer.  And even for people using Linux: almost nobody uses LC_CTYPE=*.Latin-1 nowadays, which would imply that you could no longer look at the preconv output with a pager.  When you do groff-specific encoding anyway, it's much better to encode all non-ASCII characters and not force users to adopt an obsolete locale.


While in general, i hate adding options to programs, in particular when it can be expected that they will be used rarely, i do see that an occasional need for what Brandon asks for might arise.  When picking new options, please don't forget to look at https://mandoc.bsd.lv/man/man.options.1.html - the groff/man option space is seriously crowded already, and having several programs in a single package or in two very closely related packages that all use the same option letter but each one for a different purpose isn't user-friendly at all.

Either way, i would judge this task as somewhat low-priority because the situation that you want to maintain the document source in US-ASCII (which implies there are only occasional non-ASCII characters in it, otherwise you would surely maintain the document source in UTF-8 in the first place) yet that there is a sufficient number of stray wide characters inside that you want to encode them automatically rather than just manually fixing them one by one may occasionally occur, but not all that often, i think.

Ingo Schwarze <schwarze>
Group Member
Sat 25 Jul 2020 08:02:19 PM UTC, comment #1: 

Stepping back a bit: preconv is essentially a bit of a hack to address groff's limitation of natively handling only Latin-1 input.  As long-term strategies go, addressing this limitation in core groff is a better fix than patching up the interim tool.

To keep existing pipelines working, preconv would have to exist in some form for a while, but if groff natively accepted UTF-8 input (bug #40720), preconv could turn into a simple wrapper for iconv, instead of being essentially a reimplementation of it but outputting groff-isms rather than standard character sets.  Fewer wheels reinvented.

And making groff speak UTF-8 shouldn't require wheel reinvention either.  I'm no C++ programmer, but surely the language has standard libraries to handle UTF-8, that would just need to be plugged into the appropriate places in groff's input handling.  (I say "just" in complete ignorance of how big this task actually is.)

Coders being in short supply here, it probably makes sense to devote this limited resource to the best long-term solution.

However, I have to assume preconv exists at all because writing it from scratch was once deemed substantially easier than updating groff's input handling.  So if retooling preconv remains the substantially easier task, I have no quarrel with the proposals put forth here.

(Since groff does speak Latin-1, I'm actually not sure why preconv need emit things like \['e] or \[u00E9] at all, rather than the more widely understood Latin-1 character those things represent.)

Dave <barx>
Group Member
Tue 21 Jul 2020 11:27:56 AM UTC, original submission:  

preconv is a good thing but its conversion of absolutely everything that isn't US-ASCII to inscrutable hexadecimal Unicode points means that it's not good for "offline" source document conversion in a way that would be pleasant to maintain.

I'd like:

1. A flag, possibly -s, to convert non-ASCII code points to special character escapes, e.g., \['e], where such entities exist.  The Unicode-to-glyph list in src/libs/libgroff/uniglyph.cpp` might be helpful here.

2. A flag, possibly -t, to do the same except convert only those code points documented in CSTR #54 to special character escapes, and use the old-fashioned escape form, e.g., \('e.

In the slightly longer term I'd like a Single Source of Truth for the special character escapes, one that can be macro-processed (not by groff!) into `uniglyph.cpp` and `glyphuni.cpp` as well as groff_char(7) and grops(1) (which is where I think the PostScript column of the tables in groff_char(7) today should move to).

Thoughts?

G. Branden Robinson <gbranden>
Group administrator

 

(Note: upload size limit is set to 16384 kB, after insertion of the required escape characters.)

Attach Files:
   
   
Comment:
   

No files currently attached

 

Depends on the following items: None found

Items that depend on this one: None found

 

Carbon-Copy List
  • -email is unavailable- added by schwarze (Posted a comment)
  • -email is unavailable- added by barx (Posted a comment)
  • -email is unavailable- added by gbranden (Submitted the item)
  •  

    There are 0 votes so far. Votes easily highlight which items people would like to see resolved in priority, independently of the priority of the item set by tracker managers.

    Only logged-in users can vote.

     

    No changes have been made to this item

    Back to the top

    Powered by Savane 3.13-d3ae.
    Corresponding source code