Discussion List Archives

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [ddlm-group] UTF-8 BOM

I agree with Joe that the best way to deal with character decoding and
encoding is in the I/O layer, separate from and below the lexical
analysis.  Previously, however, this group seems seen potential
performance advantage in deferring decoding until some later point,
and therefore CIF2 is currently designed with the intention of making
that possible.  (I may be missing something here, though, as I have
had to rely on reading the archives after the fact.)  The present CIF2
specification thus avoids assigning any special syntactic significance
to any non-ASCII character, such as by designating U+FEFF to be
whitespace.

All of us have been in this business long enough, I think, to know how
little performance assertions are worth without the support of good
performance tests.  Here's one data point, therefore: on my personal
workstation, using a well-tuned, pure Java decoder, UTF-8 decoding of
a 100 KB html file written in French takes about 0.2 milliseconds.
Performance will vary with hardware, file content, and decoder
implementation, but based on that model, I feel comfortable asserting
that the decoding cost for CIFs should generally be on the order of
microseconds per kilobyte.  In absolute terms, then, I don't see much
improvement to be had from avoiding some of that cost, even when
scaling to thousands of large CIFs.

Another relevant test would be the performance of a byte-based scanner
operating on the encoded byte stream, versus a character-based scanner
operating on the decoded stream.  This is a more difficult test to
perform (fairly), because it requires two separate, yet equivalent
scanners.  I'm not prepared to make such a measurement myself, but I
am also unwilling to make any assumption about what the results would
show.  My guess -- not suitable for supporting a design decision --
would be that with appropriate tuning, the two scanners' performance
would be quite similar.

Why does all that matter?  Because where there is agreement on
desirable CIF2 processor behavior, I would much rather incorporate it
into the specification than relegate it to "best practice" status.
This will yield better uniformity and fewer incompatibilities among
CIF2 implementations, which is why I raised the issue of an initial
UTF-8 BOM in the first place.  As for an internal U+FEFF, I now see
that it is important to make an explicit decision on its meaning, and
to put that, too, in the spec.  I see three viable possibilities:

1) U+FEFF after the beginning of a CIF is an ordinary character.  This
is the current status, and with this interpretation, U+FEFF may appear
in data names and whitespace-delimited data values.  Character
decoding can be deferred under this option.  This is my least
preferred of these options, but it would work.

2) U+FEFF after the beginning of a CIF is whitespace.  This runs
against the principle that non-ASCII characters should not be assigned
special syntactic significance, but it would not require giving any
such significance to any other character.  It might still be possible
to defer character decoding with this option, at the cost of a more
complicated lexical scanner.  This would be my preference.  This
interpretation yields the behavior most likely to be expected of CIFs
formed by naively concatenating multiple well-formed, CIF2-compliant
files (but see below).

3) U+FEFF after the beginning of a CIF is not allowed.  This would be
my second choice.  In the event that naive concatenation of CIFs
introduced U+FEFF into the body of a CIF, a CIF2-compliant processor
would be obliged to treat it as an error (at which point the
processor's further behavior would be outside the scope of the CIF
specification). It might still be possible to defer character decoding
with this option, at the cost of a more complicated lexical scanner.

Options (2) and (3) are both consistent with Unicode's
recommendations; option (1) is not.

Overall, I think the possibility of U+FEFF being introduced into the
body of a CIF by naive concatenation of two CIFs is a bit of a straw
man, because CIF2 introduces much deeper problems with naive
concatenation of CIFs.  Because CIF2 is not 100% backwards compatible,
a CIF formed by concatenating a CIF2 file and a CIF1 file might not be
compliant with either spec, and if it were compliant, it might not
have the same meaning as the two component CIFs taken separately.
Moreover, whether it was compliant, with which spec, and with what
meaning could depend on the order in which the files were
concatenated.

It would be possible to design CIF2 so that concatenation of two
CIF2-compliant files is always well-formed CIF2 (see option (2)
above), but that's already a break from naivete, as you have to
distinguish between CIF1 and CIF2.  I think it would be better to
simply acknowledge that with the introduction of CIF2, it is no longer
safe to combine CIFs using CIF-unaware tools.


In Summary: CIF2 should explicitly specify a particular meaning for
U+FEFF in the body of a CIF, rather than leaving any doubt as to
compliant processor behavior in this area.  CIF2 already implicitly
provides such a specification, but either the current specification
should be made explicit, or an alternative should be chosen and
specified.  In evaluating which way to proceed, it should be
recognized that quite apart from the handling of U+FEFF, CIF2 already
fails to support naively concatenating CIFs.


Best Regards,

John

P.S.: My, what a novel!  Sorry for being so long winded!  -- JCB

--
John C. Bollinger, Ph.D.
Department of Structural Biology
St. Jude Children's Research Hospital


-----Original Message-----
From: ddlm-group-bounces@iucr.org [mailto:ddlm-group-bounces@iucr.org] On Behalf Of Joe Krahn
Sent: Wednesday, May 12, 2010 5:34 PM
To: Group finalising DDLm and associated dictionaries
Subject: Re: [ddlm-group] UTF-8 BOM. .

In general, CIF does not directly deal with encoding. It should be
possible to allow a low-level I/O library to deal with all encoding
issues. Therefore, supporting non-standard stream interpretation should
be avoided.

It is a good practical idea to allow mid-stream BOMs by interpreting
them as character 0xFEFF, and allow it as whitespace, with a warning. It
should not be a required feature, because it is non-standard, and only
exists in UTF-8 for backwards-compatibility. Eventually, conforming I/O
libraries will interpret them as invalid. Ideally, text concatenation
software will become BOM-aware.

Interpret mid-stream BOMs and allowing mixed encodings is a major hack,
and impractical for systems that deal with encoding at the I/O level. It
is reasonable to allow it as an non-standard extension, but should
always give a warning so that people realize that such files are likely
to be broken elsewhere.

In summary: Standard CIF2 needs to support standard UTF-8 BOMs at the
beginning of a file. Anything else should be considered a non-standard
extension. For practical reasons, CIF2 parsers should be encouraged but
not required to allow mid-stream UTF-8 0xFEFF as whitespace.

Joe

Herbert J. Bernstein wrote:
> Dear Colleagues,
> 
>    While it is certainly prudent to tell people to either write a pure
> UTF-8 file with no BOM or to prefix it with a BOM, and that is
> home a compliant CIF writer should work, it is not practical to
> insist the CIF readers should reject embedded BOMs.  Indeed, the
> URL cited by John does not tell you they are illegal, but that
> you should treat them as a zero width non-breaking space.
> 
>    The reason we cannot insist on readers demanding that BOMs occur
> at the beginning is that users may concatenate whole CIF or
> build one CIF out of fragments of text, and this will very likely
> result in embedded BOMs and possibly switches in encodings.  If
> we fail to handle the BOMs were are much more likely to garble
> such files.  I strongly recommend the approach in my prior
> message -- recognize BOMs are all times.
> 
>    Regards,
>      Herbert
> 
> =====================================================
>   Herbert J. Bernstein, Professor of Computer Science
>     Dowling College, Kramer Science Center, KSC 121
>          Idle Hour Blvd, Oakdale, NY, 11769
> 
>                   +1-631-244-3035
>                   yaya@dowling.edu
> =====================================================
> 
> On Tue, 11 May 2010, Bollinger, John C wrote:
> 
>> Dear Colleagues,
>>
>> I think CIF processor behavior such as Herb describes would be
>> outstanding, and I commend Herb for his dedication to providing such
>> capable and robust software.  I do disagree about one of his specific
>> points, however:
>>
>>> The
>>> minimum to do with any BOM is:
>> [...]
>>
>>>   1.  Accept it at any point in a character stream.
>> It would be both unconventional and programmatically inconvenient to
>> give special significance to U+FEFF anywhere other than at the very
>> beginning of a file.  The Unicode consortium in fact addresses this exact
>> question in its FAQ: http://www.unicode.org/faq/utf_bom.html#bom6.
>> Although the Unicode's comments do allow for protocol-specific support
>> for accepting U+FEFF as a BOM other than at the beginning of the stream,
>> I see little advantage to adding such a complication to the CIF2
>> specifications.
>>
>> This all expands the scope of the topic far beyond what I had intended,
>> however.  I think it is perhaps useful to recognize at this point,
>> therefore, that the CIF2 language specification and the behavior of CIF2
>> processors are separate questions.  This group has already decided that
>> files compliant with CIF 2.0 are encoded in UTF-8, period.  I do not want
>> to reopen that debate.  On the other hand, that in no way prevents CIF
>> processors from -- as an extension -- recognizing and handling putative
>> CIFs that violate the spec by employing character encodings different
>> from UTF-8.  That sort of thing is generally heralded as beneficial for
>> ease of use, and it is consistent with the good design principle of being
>> relaxed about inputs but strict about outputs.  (And in that vein I would
>> hope that any CIF 2.0 writer's normal behavior would be to encode in
>> UTF-8.)
>>
>> My suggestion is slightly different, as I hope this restatement will
>> show: *in light of the fact that spec-compliant CIF2 files are encoded in
>> UTF-8*, I suggest that the spec allow a file beginning with a UTF-8 BOM
>> to be spec-compliant (subject to the compliance of the rest of the
>> contents).  Like Herb, I intend that my parsers will accept such CIFs
>> whether they strictly comply with the spec or not, but the question is
>> whether accepting such files should be a compliance requirement or an
>> extension.  Either way, I think it will be valuable to document this
>> decision in the spec, if only to draw attention to the issue.
>>
>>
>> Best Regards,
>>
>> John
>> --
>> John C. Bollinger, Ph.D.
>> Department of Structural Biology
>> St. Jude Children's Research Hospital
>>
>>
>> Email Disclaimer:  www.stjude.org/emaildisclaimer
>>
>> _______________________________________________
>> ddlm-group mailing list
>> ddlm-group@iucr.org
>> http://scripts.iucr.org/mailman/listinfo/ddlm-group
>>
> _______________________________________________
> ddlm-group mailing list
> ddlm-group@iucr.org
> http://scripts.iucr.org/mailman/listinfo/ddlm-group
> 

_______________________________________________
ddlm-group mailing list
ddlm-group@iucr.org
http://scripts.iucr.org/mailman/listinfo/ddlm-group


_______________________________________________
ddlm-group mailing list
ddlm-group@iucr.org
http://scripts.iucr.org/mailman/listinfo/ddlm-group

Reply to: [list | sender only]
International Union of Crystallography

Scientific Union Member of the International Council for Science (admitted 1947). Member of CODATA, the ICSU Committee on Data. Member of ICSTI, the International Council for Scientific and Technical Information. Partner with UNESCO, the United Nations Educational, Scientific and Cultural Organization in the International Year of Crystallography 2014.

ICSU Scientific Freedom Policy

The IUCr observes the basic policy of non-discrimination and affirms the right and freedom of scientists to associate in international scientific activity without regard to such factors as ethnic origin, religion, citizenship, language, political stance, gender, sex or age, in accordance with the Statutes of the International Council for Science.