SYSTEM vs. PUBLIC

[Razvan]

Hi !

Sometimes ago I posted a question regarding the use of the PUBLIC vs SYSTEM declaration in an XML file. After reading the initial answers and after gaining a little more experience with XML I came up with the following conclusions:

SYSTEM declaration can be used to specify a file on the local file system like:

<!DOCTYPE RootElement SYSTEM "C:\validate.dtd">

The problem with this approach is that if the file is made public the path specified on the local file system will not have any meaning any more. Even if the path specified in the SYSTEM declaration *is* a URL:

<!DOCTYPE RootElement SYSTEM "http://www.mihaiu.name/validate.dtd">

the parser might be unable to retrieve the DTD file if the system is not connected to the Internet.

The PUBLIC declaration constitutes a partial solution to this problem. The string contained in a PUBLIC declaration is not an URL but an URN (Uniform Resource Name). A URN does not pinpoint the precise location of the resource, but only clearly specify its name. The *parser* of the document must be smart enough to be able to generate a URL from a URN using some internal logic.

Example of a PUBLIC declaration:

<!DOCTYPE RootElement PUBLIC "mihaiu/validate.dtd"
SYSTEM "http://www.mihaiu.name/validate.dtd">

In this case, a custom parser that already has a catalogue of DTDs published by mihaiu can generate a URL from the PUBLIC declaration. The generated URL can look like

c:\DTDs\validate.dtd

There is no standard way to convert a URN to a URL, so, if this conversion fails because the parser does not contain the internal logic to perform such a conversion (or for whatever other reason) the parser will attempt to use the SYSTEM declaration which in this case resolves to

http://www.mihaiu.name/validate.dtd

Important observation:
Since there is no standard way to generate a URL from a URN the PUBLIC declarations can only be useful for customized parsers !!! (e.g. they are not useful for general purpose parsers like Xerces)

DID I GOT IT RIGHT ?

Regards,
Razvan

www.mihaiu.name

 

[Johannes Koch]

Hi !

Sometimes ago I posted a question regarding the use of the PUBLIC vs SYSTEM declaration in an XML file. After reading the initial answers and after gaining a little more experience with XML I came up with the following conclusions:

SYSTEM declaration can be used to specify a file on the local file system like:

<!DOCTYPE RootElement SYSTEM "C:\validate.dtd">

As far as I know, the system identifier must be a URL (or URI), so it must be something like "file://C|/validate.dtd".

The problem with this approach is that if the file is made public the path specified on the local file system will not have any meaning any more.
Right.

Even if the path specified in the SYSTEM declaration *is* a

.... HTTP ...

URL:

<!DOCTYPE RootElement SYSTEM "http://www.mihaiu.name/validate.dtd">

the parser might be unable to retrieve the DTD file if the system is not connected to the Internet.
Yes.

The PUBLIC declaration constitutes a partial solution to this problem. The string contained in a PUBLIC declaration is not an URL but an URN (Uniform Resource Name).

No, it's not a URN, it's an FPI (formal public identifier), although it acts somehow like a URN. To map FPIs to system identifiers, the parser may use a catalog (look for XML Catalog or OASIS Open Catalog).

DID I GOT IT RIGHT ?

Not really.

 

[David Carlisle]

As far as I know, the system identifier must be a URL (or URI), so it must be something like "file://C|/validate.dtd".

It's correct that it has to be a URI, although the mapping (usually) suggested between URIs and windows paths is file:///c/validate.dtd rather than the form with | (which was used by some early netscapes)

No, it's not a URN, it's an FPI (formal public identifier), although it acts somehow like a URN. To map FPIs to system identifiers, the parser may use a catalog (look for XML Catalog or OASIS Open Catalog).

Although most systems (HTML, Docbook, ...) that use a PUBLIC ID do use FPI syntax, that is a refection of the SGML ancestory of these systems.

XML just defines this to be a more or less arbitrary string (of ASCII printing characters), and the XML (unlike SGML) has no way of specifying that the PUBLIC identifier has any specific syntax, it just enforces a set of characters:

[13] PubidChar ::= #x20 | #xD | #xA | [a-zA-Z0-9] | [-'+,./:=?;!*#@$_%]

David

 

[Razvan]

Please tell me what the difference between URN and FIPS is. Can you pinpoint me to a web site where I can find out more about FIPS and especially about the subtle differences between URN and FIPS?

By the way: FIPS are URIs ?

How about the rest ? Ok: it's not URN, it's FIPS. Just replace URN with FIPS. The rest of my essay is correct ?

Regards,
Razvan

 

[Johannes Koch]

Please tell me what the difference between URN and FIPS is.

FPIs (= Formal Public Identifiers), not FIPS.

Can you
pinpoint me to a web site where I can find out more about FIPS and especially about the subtle differences between URN and FIPS?

Use the search engine of your choice. E.g. google search for "formal

By the way: FIPS are URIs ?

No. Read RFC 2396 for the definition of URI.

How about the rest ? Ok: it's not URN, it's FIPS. Just replace URN with FIPS. The rest of my essay is correct ?

No. Read about FPI, URN, the difference between windows file path and URL, catalogs, entity resolver, ...

 

[Arjun Ray]

SYSTEM declaration can be used to specify a file on the local file system

[Note: _identifier_, not _declaration_]

Yes, in the sense of "original intent" going back to SGML, and no, because XML has repurposed SYSTEM into a slot for URIs.

The PUBLIC declaration constitutes a partial solution to this problem.

Arguably, this may be what the XML spec intends.

The string contained in a PUBLIC declaration is not an URL but an URN (Uniform Resource Name).

Not really. In XML, a PUBLIC identifier is what SGML calls a _minimum literal_, an arbitrary string with characters limited to a certain small set.

Now, it so happens that PUBLIC identifiers in *SGML* (not XML!) are very much like URNs, inasmuch as they share the properties of persistence and uniqueness. XML usage has inherited this understanding in some quarters, but there is no formal basis for this in the XML spec.

DID I GOT IT RIGHT ?

Only in part. You are trying to make sense of something that has been screwed up beyond any reasonable hope of recovery. The critical blunder was the repurposing of SYSTEM identifiers for URIs. Unfortunately, URIs don't fit into the PUBLIC/SYSTEM dichotomy of SGML at all, but the catchphrase, "Cool URIs don't change", is a good indication that URIs are really more *useful* in their PUBLIC than in their SYSTEM aspect.

Note that the XML spec doesn't *define* the PUBLIC and SYSTEM keywords, i.e. explain what they mean and thus why they're different (and both there to begin with). One could fill this gap from SGML, but again, there is no formal basis for this in the XML spec.

In SGML, PUBLIC means "well-known, widely understood/accepted" while SYSTEM means "private, local, custom, homegrown, proprietary". For those who have Goldfarb's SGML Handbook, there is an extensive, and IMHO clear, exposition of this on p.378-9. He writes, inter alia,

A _public identifier_ is a name that is intended to be meaningful across systems and different user environments.

A _system identifier_ is system-specific information that enables the entity manager component of an SGMl system to locate the file or the memory location or the pointer within a file where the entity can be found [...] a system identifier could be an invocation of a program that controls access to an entity that is being identified.

The system identifier itself need not be the full storage identifier; it is just a method of expressing information that the entity manager can use to determine the storage identifier [...] In that regard, it would be very sensible for an implementation to devise a defaulting scheme in which the storage identifier could be determined from the entity name alone. SGML encourages this by providing syntactically that the keyword SYSTEM can be specified for an external identifier without actually specifying a system identifier at all.

In other words, you have complete freedom to decide what your SYSTEM identifiers "mean". And, in general, system identifiers should *not* be used in documents meant for exchange, because there is no expectation that anyone except you could make sense of them. At a pinch, you could leave just the SYSTEM keyword in, hoping that the other guy has as sensible system of defaults in his catalogs as you.

Thus, XML goofed in two ways to take this useful functionality (how you organize your own local system) away from you - by mandating the presence of the system identifier, and further constraining its form to URIs.

Take it from there.