Agree with the rewording for the 'isolated ES or WSP*' cases.

I'm not convinced that we need to allow ES and WSP* in isolation for separators. When I present on DFDL, I get asked when to use a separator or terminator. My answer is use a separator when the delimiter is always the same between occurrences. which implies it is a property of the sequence rather than each element. Allowing ES or WSP* is breaking that, to my mind. I'm also not sure what effect it has on separator suppression, which is complicated enough as it is. So I'd prefer to leave separator wording as it is.

I'd rather not introduce the concept of scanning for initiators. The parser does not really scan for initiators, it expects to find an initiator at the current offset. The initiatedContent property was added to allow a) the schema to be checked to ensure all children had an initiator, and b) initiators to be discriminators. (Scanning for separator/terminator is different as the parser really does have to scan through bytes to find a separator/terminator.)

Regards

Steve Hanson

IBM Hybrid Integration, Hursley, UK
Architect, IBM DFDL
Co-Chair, OGF DFDL Working Group
smh@uk.ibm.com
tel:+44-1962-815848
mob:+44-7717-378890
Note: I work Tuesday to Friday

From: Mike Beckerle <mbeckerle.dfdl@gmail.com>
To: Steve Hanson <smh@uk.ibm.com>
Cc: DFDL-WG <dfdl-wg@ogf.org>
Date: 09/10/2018 18:25
Subject: Re: [DFDL-WG] Clarification needed: sequence terminator that exists or not depending on expression

Comments inline.
...mikeb

On Tue, Oct 9, 2018 at 9:41 AM Steve Hanson <smh@uk.ibm.com> wrote:
"%ES;%ES;" is already disallowed, as ES can only appear once - see the entity syntax table.

"%ES; %ES;" is also disallowed, it contravenes the first sentence "ES must not appear as the only DFDL string literal in the property." It appears twice, but it is still the only DFDL string literal :) The wording is clearly ambiguous as we interpreted it differently.

Suggest rewording as: "Neither '%ES;' nor '%WSP*;' may appear as an isolated string literal in the property value, or in the value returned from an expression when scanning for delimiters."

Note that IBM DFDL has not yet implemented the erratum (2.148) that allows ES to appear anywhere other then dfdl:nilvalue. (All started from this public comment https://redmine.ogf.org/boards/15/topics/40)

IBM has also encountered this type of "variable-length-with-max" string. I'm sure I raised it in the WG a long time ago, and we discussed (and presumably rejected) whether it should be a new lengthKind, eg "delimitedMax", for convenience. Can't find anything in my email logs though. And not sure what we did to model it ?? My memory could be playing tricks.

I don't want to add a length kind for this. I want to be able to use delimiters both when scanning for terminating markup, and when not doing so, and have what is allowed in terminating markup be different for the two cases, based on whether lengthKind='delimited' applies anywhere the delimiters are in scope.

We already have this language in the DFDL spec. i.e., designed to work this way, it's just not complete and consistent.

Whatever we decide, each of initiator, terminator and separator need to be considered separately. Note that ES is currently allowed (with stated restrictions) for initiator and terminator only, not for separator - which makes sense to me but is contrary to 2.148 ??
Also must be wary of EVDP.

And NVDP also.

Separators can also be used when NOT scanning for terminating markup. E.g., a sequence of 10 fixed-length strings can have comma separators. No scanning is used for them, as each child is just 10 long exactly, and then the separator must be found. In this case having %ES; as one of the string literals just means there may or may not be found any of the separators, i.e., they are optional.

I went and re-read 2.148, the trackers for the public comment, etc.

We just need a crisp and complete definition of what scanning for delimiters means.

There are two cases:

Case 1: Scanning for initiators
We are scanning for an initiator when initiatedContent="yes" and we are parsing the
* children of a choice group
* children of an unordered sequence group
* children of a sequence group having floating="yes"
When scanning for an initiator, an initiator must be defined and in-effect.

This means when the child (per above) is
* an element where the value can be empty, EVDP must be initiator or both along with an initiator being defined.
* a nillable element, NVDP must be initiator or both along with an an initiator being defined.
This whole EVDP/NVDP discussion is probably unnecessary if we just say "initiator must be in-effect".

In other cases we're not scanning for initiators.

Case 2: Scanning for length

We are scanning for length when lengthKind='delimited' and we are parsing an element.

Section 12.3.2 describes this, though it doesn't discuss details of determining length of the nil representation. This section could be improved, but I'm not really worried about that right now.

So scanning for delimiters is either scanning for initiators or scanning for length. In that case, none of the in-scope terminating delimiters can be %ES; nor %WSP*; in isolation.

So this suggests in summary:
* section 12.2 phrasing of constraints on %ES; and WSP* must be improved to be clearer and less ambiguous for initiator and terminator.
* section 14.2 (definition of separator property) needs updating to match that of terminator. The terminator property specifies both ES; and WSP* entities are not allowed if scanning for delimiters. Separator needs to be the same.
* Section 12.2 description of initiator needs to say that %ES; and %WSP* in isolation are not allowed if scanning for initiators.
* A new 12.2 sub-section should be added that defines "scanning for initiators", and section should be referenced from the description of initiator property.

Regards

Steve Hanson
IBM Hybrid Integration, Hursley, UK
Architect, IBM DFDL
Co-Chair, OGF DFDL Working Group
smh@uk.ibm.com
tel:+44-1962-815848
mob:+44-7717-378890
Note: I work Tuesday to Friday

From: Mike Beckerle <mbeckerle.dfdl@gmail.com>
To: DFDL-WG <dfdl-wg@ogf.org>
Date: 01/10/2018 20:31
Subject: [DFDL-WG] Clarification needed: sequence terminator that exists or not depending on expression
Sent by: "dfdl-wg" <dfdl-wg-bounces@ogf.org>

Consider the following:

<element name="value" type="xs:string" ...../>
<sequence dfdl:terminator="{ if (fn:string-length(./value) eq 32) then '%ES;' else '%NUL;' }"/>

This is used to add a NUL at the end of a string, if the string length is less than the max length of 32. This comes up often in fixed length or variable-length-with-max data we've seen. I've put this terminator on a separate sequence after the element to emphasize that we're not scanning for terminating markup here. This has nothing to do with lengthKind 'delimited'.

However, the DFDL spec says (for terminator property)

· ES must not appear as the only DFDL string literal in the property. It can only appear as a member of a list.

· Neither the ES entity nor the WSP* entity may appear on their own as one of the string literals in the list when the parser is determining the length of a component by scanning for delimiters.

The second bullet doesn't apply to my example.

Re: first bullet, I think my terminator expression is illegal... because the '%ES;' is a list of literals containing ES as the only DFDL string literal.

But this is a really flawed constraint, as "%ES;%ES;" and "%ES; %ES;" both skirt the constraint, but mean the same thing as just "%ES;" which is illegal.

So, if we don't want to allow these hack workarounds, we need a statement that says runs of %ES; adjacent mean the same thing as one %ES;, and that more than one identical-meaning delimiter specified in a list of string literals means the same as just one. Or we can make these hack workarounds illegal.

However, why are we disallowing these?

The above construct in my example is very useful, and really hard to work around unless we can have a terminator that is '%ES;' as the only string literal. Actually I have no work around for this really. I am guessing I could come up with something, but the various things I've guessed at don't pan out, or prevent the string named 'value' above from being modeled as a simple type.

I know we don't want lengthKind='delimited' with terminator="%ES;" as that is most likely just a schema-definition error, but when we're not dealing with a lengthKind, we really do seem to need to specify situations where conditionally the terminator region will be empty.

So I think we need to do:
1) clarify that %ES; cannot be used in combination with any other character or entity as a member of a list of string literals.
1a) At the same time I would also disallow combinations of WSP* that are misleading and unnecessary i.e., disallow %WSP*; adjacent to any other WSP, WSP+, or WSP*.
2) clarify that the constraint that %ES; for terminator and separator cannot appear as the only string literal in a list of string literals... applies only when the parser is determining the length of a component by scanning for delimiters. This is just rephrasing the two bullets above so the clause about scanning applies to both, not just the second.

I believe this preserves the intent that when lengthKind="delimited" and we are scanning for delimiters, there must be *some* delimiter that is potentially not zero length. You still have to cope with the possible match being zero length due to %ES; being in the list of terminating markup, or WSP* similarly, with no whitespace found. But the notion that there is NO scanning to be done can't happen. That is, the notion that the schema specifies lengthKind delimited, but also specifies no delimiters at all, is still ruled out.

Comments?

Mike Beckerle | OGF DFDL Workgroup Co-Chair | Tresys Technology | www.tresys.com
Please note: Contributions to the DFDL Workgroup's email discussions are subject to the OGF Intellectual Property Policy -- dfdl-wg mailing list dfdl-wg@ogf.org https://www.ogf.org/mailman/listinfo/dfdl-wg

Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number 741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU

Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number 741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU