Re: [tdwg-guid] LSID metadata persistence (or lack thereof)

14 Jul 2007


      Folks,
Thanks much to all of you who replied to my post. All the posts were 
really relevant to our discussion.
Before we go ahead, however, let us stop for a minute to try and 
summarize the points we agree upon and the points in which there is 
still significant controversy.
I believe that we reached consensus in the following issues:
1) We do agree that *LSID metadata is not required to be persistent* 
(i.e. clients cannot assume it is immutable). See note [1].
2) We should not force XML representations of data to be byte identical 
just to return that in the LSID getData() call. We must find another way 
to fulfill this requirement.
3) We should not try to return something in the LSID getData() call just 
for the sake of it. We shouldn't for example return the bare scientific 
name of a species in the getData() call just because that can be 
immutable and thus fulfill the requirement from the LSID spec. This is 
counterproductive because the name itself is in the metadata already and 
no client would gain anything from calling getData() in this case.
We have also raised new issues that may be worth discussing (in 
their own separate thread if possible):
4) We "may" bend the immutability rule of LSID getData() to our benefit 
and accept data that is not byte stream identical, but only 
"semantically" identical (depending on content type maybe). If we do 
this, we may use the LSID getData() call more effectively to identify 
real datasets such as matrices, identification keys, etc.
5) As Brian pointed out, we may need to revisit what we call data and 
metadata. We have been using the LSID getMetadata() call to return what 
some people may call data (taxon names, specimens, collections). And we 
forgot completely that there may be other kinds of data out there that 
may be returned in the getData() call and that those still need metadata 
to describe them. I think this may be worth discussing in a separate thread.
Did I leave anything out? If so, please let us know by replying to 
my post and adding a short entry to either list above.
Cheers,
Ricardo
Notes:
-------
[1] Matt may disagree with me here, but my point is that we can't force 
all authorities (i.e. data providers) to keep perfect archives of all 
versions of their databases given a heterogeneous and distributed 
environment we operate in. While some may want to provide this feature, 
other providers may not want or be able to.
Richard Pyle wrote:
...
It seems to me that there is a third method to resolving the problem:
When we want to identify an object that is itself digital in nature (e.g., a
database record, or a binary data file such as a PDF, JPG, ASCII, Unicode,
or whatever), we resolve said binary object via getData().  If, for some
reason, we change the exact bit-sequence of that digital/binary object
(e.g., color-correct an image, change a text string from ASII to Unicode, or
whatever...), we assign a new LSID to it (whether that "new" LSID differs
from the "old" LSID only via the optional "Revision" part of the LSID, or
via a new Object Identification part, is a topic for another debate).
When we want to identify an object that does not itself have a digital
manifestation -- like a physical object (e.g., specimen or a particular
printed copy of a publication) or an abstract/conceptual object (e.g., a
taxon name, a taxon concept, a geographica place, or a cited publication) --
then we return *nothing* in response to getData(), and we treat all the
attributes of said physical/abstract/conceptual object of interest to us as
metadata.
If there are cases where certain metadata elements of an object without an
inherent digital existence need to persists (and there are), yet we also
want to allow modifications to metadata elements without the need to
generate new identifiers for the underlying object (and we do) -- then we
deal with those within our own community via adopted standards and best
practices.
I would disagree strongly with bending the existing LSID standard, and would
just as strongly favor working within its existing framework (which, I
think, we can).  I would also disagree with the practice of embedding XML
documents as "data" for an LSID, unless the LSID is intended to represent
the XML document itself (in which case there might be a different LSID to
represent the database record that was used to generate the XML document;
and yet another LSID to represent the abstract concept that the database
record was created to represent -- like a taxon name, for example).
If we want to use LSIDs to pass around XML packages (that are not rendered
as RDF) about abstract objects (e.g., taxon names), why doesn't our
community define within our semantic vocabulary something along the lines of
"TCS_XML", which can be established as a standard metadata component for
LSIDs assigned to taxon concepts (i.e., abstract objects, identified by
"data-less" LSIDs).  The exact bytestream of the content of that metadata
element can change, without changing its canonical rendering.
I'm beginning to suspect (strongly) that I am completely missing some
fundamental point here -- and perhaps is is the same point that underlies
the apparent antagonism towards LSIDs in general (which I do not yet share).
But I am fairly certain we are dealing with some level of miscommunication
here.
Aloha,
Rich
...
-----Original Message-----
From: tdwg-guid-bounces@lists.tdwg.org 
[mailto:tdwg-guid-bounces@lists.tdwg.org] On Behalf Of P. 
Bryan Heidorn
Sent: Friday, July 13, 2007 12:48 PM
To: Dave Vieglais
Cc: tdwg-guid@lists.tdwg.org
Subject: Re: [tdwg-guid] LSID metadata persistence (or lack 
thereof)[Scanned]
There seems to be two methods to resolving this problem.
One is to change the LSID definitions to allow semantic 
equivalence in the data and not require exact bit stream equivalence.
The other option is to change the data representation so that 
it is "easily" reduced to a repeatable canonical form. For 
example, it is almost as easy as saying where XML ordering 
does not specify order of elements, elements will be ordered 
alphabetically. Seems stupid but it almost works.. except 
where you have repeating elements with the same element name 
where it does not work.
It seems a little odd to bend the standards for the data 
being delivered to fit the requirement of the LSID spec. In 
theory, the other standard developers who set the data being 
delivered did not fix order because it did not matter.
This is different from Chuck's observation that the semantics 
of the element within some of the standards are 
insufficiently specified.  
So, what we mean is a darwin mode species name is just a 
string and nothing more now.
--Bryan
On Jul 13, 2007, at 5:18 PM, Dave Vieglais wrote:
...
I think we are all in agreement that the data and metadata
referenced
...
by an LSID remains unchanged (in the case of the metadata, semantic 
equivalence is a requirement for reasons such as outlined
by Matt).
...
My question is to do purely with the data that an LSID references 
through the getData() operation.  The form of that data could be 
anything really - an encrypted byte stream, digital image,
Open Office
...
document, spreadsheet, xml document...
We all know that the same data can be represented many ways
that are
...
logically, semantically and functionally equivalent yet form a 
different set of bytes when serialized.  Data expressed in
XML is one
...
example (is <a/> = <a /> = <a></a> ?).  A pallet based image is 
another - the order of colors in the palette may be
changed, and the
...
pixel values adjusted to match the new palette order, but
the image is
...
still the same. There are many more simple examples that can be 
constructed that violate the unchanged bytes rule but for all 
practical and functional purposes the data are unchanged.
As mentioned previously, enforcing and implementing the unchanged 
bytes rule is not challenging. It is however quite different from 
stating that the data are returned unchanged.  It is this
that I, and
...
I'm sure a lot of other implementors would appreciate consensus on.
Dave V.
On Jul 14, 2007, at 09:20, Matthew Jones wrote:
...
In terms of the metadata returned from an LSID, or any
other digital
...
...
identifier, there are definite cases where metadata must be 
semantically persistent in order to preserve the utility
of data and
...
...
accuracy of scientific results.
As a trivial example, given a set of observations
collected at time
...
...
t, one can represent the data for those observations in
dataset D and
...
...
the metadata for the dataset, including the time value t, in a 
metadata document M.  In a later event, it is discovered
that t was
...
...
entered incorrectly, and needs to be adjusted, creating metadata 
document M'. That M and M' are not congruent is critical knowledge 
when analyzing data from D with data from another dataset D2.  In 
other words, because there is no true distinction between data and 
metadata (any given piece of information can be stored in either 
location), a proper archive must be able to distinguish
any changes
...
...
in the data and any changes in the metadata.
That said, there are some metadata that could change with
little or
...
...
no impact on data interpretation (e.g., the spelling of
the street on
...
...
which Technician Tom gets his snailmail).  But at the current time 
its impossible to distinguish this kind of metadata from the 
important kind in the general case of the existing
metadata standards
...
...
in use (e.g., FGDC BDP, ISO 19115, EML, GML, etc).
Our process in the KNB/SEEK/NCEAS and other ecological
data archives
...
...
is to give persistent identifiers to both data objects and
metadata
...
...
objects, and provide new identifiers when either changes.
Matt
Dave Vieglais wrote:
...
Hi Bob,
Just because a standard is published does not mean that it is 
practical.  Requiring that a set of bytes referenced by
an LSID are
...
...
...
unchanged has a lot of implications with respect to the 
implementation of data services.  For example, if it is agreed to 
abide by the rule that the blob referenced by an LSID remains 
forever unchanged, then that implies that the data
provider stores
...
...
...
the data as a blob, rather than risking the process of 
reconstructing on the fly from some database, especially for the 
example of data expressed in XML where functionally identical 
objects (constructed using different DOM libraries for
example) are
...
...
...
not identical blobs.
Asserting that two instances of an object with the same LSID are 
semantically equivalent is a vastly more complicated
processes than
...
...
...
asserting that the canonical representation of those
instances are
...
...
...
identical.  Generally there can be defined a simple set of 
guidelines for constructing the canonical form of an
object (eg. for
...
...
...
xml http:www.w3.org/TR/xml-c14n ) whereas asserting semantic 
equivalence is an ongoing topic of research.
Requiring identical blobs is certainly possible, but
people need to
...
...
...
be aware of the implications of such a requirement in the early 
stages of designing a system to support such a specification.  My 
preference for the canonical form relaxes the implementation 
requirements considerably whilst still maintaining the
integrity of
...
...
...
the data and the intent of the LSID.
regards,
  Dave V.
On Jul 14, 2007, at 08:08, Bob Morris wrote:
...
This entire discussion confuses me. The LSID standard is
published.
...
...
...
...
Why is there a discussion of what an LSID should be? The
standard
...
...
...
...
requires that the data, as defined by the return of
getData,  to be
...
...
...
...
identical for all resolutions of the LSID. From page 9
of the LSID
...
...
...
...
spec:
" bytes getData (LSID lsid)
bytes getDataByRange (LSID lsid, integer start, integer length) 
Metadata_response getMetadata (LSID lsid, string[]
accepted_formats)
Metadata_response getMetadataSubset (LSID lsid, string[] 
accepted_formats, string selector) The data retrieval
services may
...
...
...
...
implement all of the methods, or only methods for
retrieving data,
...
...
...
...
or only methods for retrieving associated metadata.
The same LSID named data object must be resolved always
to the same
...
...
...
...
set of bytes. Therefore, all of the data retrieval
services return
...
...
...
...
the same results for the same LSID. The user has, however, the 
choice of which one of these to utilize depending on its
location,
...
...
...
...
known quality of service and other attributes. With
metadata, the
...
...
...
...
situation is different. Each data retrieval service can provide 
different metadata for the same LSID."
This doesn't seem very ambiguous to me, and doesn't have
anything
...
...
...
...
to do with imperfect storage of data or anything else about the 
physical or electronic world. If two calls to getData() with the 
same argument on two occasions to possibly two different
resolution
...
...
...
...
services do not yield the same set of bytes, then one or
the other
...
...
...
...
or both of those is not executing a compliant service response. 
Unless this discussion is really "Shall we call something other 
than the return of getData by the term 'data associated with the 
LSID?' there seems to be nothing to discuss.
Bob
On 7/13/07, Paul Kirk p.kirk@cabi.org wrote:
>
> In an imperfect world there is no such thing as an 'identical- 
> byte-stream'
> because the technology we use is imperfect ... the disk 
> controllers which manage our bytes and the disk we use to store 
> our bytes have recognized error rates. Perhaps I'm 
>
being a pedant
...
...
...
...
> in the above analysis but I was almost persuaded that 
>
except for
...
...
...
...
> digital objects (images,
> sounds) which can
> be data all other 'things' (names, specimen accession 
>
numbers) had
...
...
...
...
> to be metadata. This to me makes no sense in the real but 
> imperfect world we live in. An LSID assigned to a name 
>
(e.g. Homo
...
...
...
...
> sapiens) is assigned to the name as data, not metadata. What is 
> 'identical' here it that if the spelling has to change for any 
> reason the new spelling gets a new LSID and the now incorrect 
> spelling gets deprecated (but is still resolvable) with 
>
a pointer
...
...
...
...
> to the correct spelling/LSID in the metadata.
>
> OK?
>
> Paul
>
>  ________________________________
>  From: tdwg-guid-bounces@lists.tdwg.org on behalf of 
>
Chuck Miller
...
...
...
...
> Sent: Fri 13/07/2007 19:03
> To: Dave Vieglais
> Cc: tdwg-guid@lists.tdwg.org
> Subject: RE: [tdwg-guid] LSID metadata persistence (or lack 
> thereof)[Scanned]
>
>
>
>
> Dave,
> What you say is true.  But, I think we already have too many 
> variations, subtleties, and reinterpretations which are 
>
endlessly
...
...
...
...
> debated.
>
> The LSID standard would be simple, clear and consistent 
>
if we used
...
...
...
...
> the identical-byte-stream definition.  The LSID would 
>
uniquely tag
...
...
...
...
> a persistent byte stream. A persistent byte stream is 
>
always the
...
...
...
...
> same thing without any further explanation or clarification.
>
> The provider of an LSID byte-stream would need to commit to 
> keeping that byte-stream persistent and not represent it in 
> multiple ways, even though technically they could.  If 
>
they can't
...
...
...
...
> commit to that, then it can't be an LSID byte-stream.
>
> And in the name of simplicity and clarity, if they had 
>
to provide
...
...
...
...
> different byte-stream representations then they would have to 
> assign a different LSID to each and use "SameAs" metadata.
>
> Chuck
>
> -----Original Message-----
> From: Dave Vieglais [mailto:vieglais@ku.edu]
> Sent: Friday, July 13, 2007 12:42 PM
> To: Chuck Miller
> Cc: Ricardo Pereira; tdwg-guid@lists.tdwg.org
> Subject: Re: [tdwg-guid] LSID metadata persistence (or lack
> thereof)
>
> Hi Ricardo, Chuck,
> Asserting that the byte stream returned as data 
>
associated with an
...
...
...
...
> LSID should never change is perhaps a bit confusing from a 
> programmatic view.  There are for example many ways to 
>
represent
...
...
...
...
> data in xml that are identical from an information 
>
content point
...
...
...
...
> of view, but the byte streams could be very different.
>
> Perhaps it might be better to state something like "the 
>
canonical
...
...
...
...
> representation of the data associated with an LSID must not 
> change", or something to that effect?
>
> Dave V.
>
> On Jul 14, 2007, at 05:29, Chuck Miller wrote:
>
>               
>> Ricardo,
>>
>> Looking at this definition: "Persistence of LSID 
>>
Data: The data
...
...
...
...
>> associated with an LSID (i.e, the byte stream returned by the
>>                 
> LSID
>               
>> getData call) must never change"
>>
>>
>>
>> Perhaps this is a more straightforward way to conceive
>>                 
> LSIDs.  The
>               
>> LSID goes with a byte stream.  It's that byte stream that
>>                 
> must stay
>               
>> the same.  So, if there is a byte stream associated with a 
>> collection that needs to stay the same, then whatever 
>>
that byte
...
...
...
...
>> stream happens to be is the data that gets an LSID assigned
>>                 
> to it.
>               
>> That sure seems a clearer definition of what is data 
>>
and what is
...
...
...
...
>> metadata, rather than the issue of primary object and 
>>
all that.
...
...
...
...
>>
>> So we can create a new definition in the context of LSIDs:  
>>                 
> Data is
>               
>> a byte stream that is persistent, never changes and 
>>
can have an
...
...
...
...
>> LSID.  Metadata is a byte stream is non-persistent, 
>>
might change
...
...
...
...
>> and is only associated with an LSID.
>>
>>
>>
>> The institution who assigns an LSID can make their 
>>
own decision
...
...
...
...
>> about whether the byte stream being provided is persistent or
>>                 
> non-
>               
>> persistent.  By assigning an LSID to any byte stream,
>>                 
> whatever it
>               
>> is, the institution is declaring it to be data and persistent.
>>
>>
>>
>> So, in the example given of an observation record with a 
>> determination that needs to remain fixed and unchanged, by 
>> assigning an LSID to that observation+determination 
>>
it would be
...
...
...
...
>> "declared to be data" and unchangeable.  A different
>>                 
> determination
>               
>> would then be different data with a different LSID.  
>>
That would
...
...
...
...
>> provide a solution for those who want to employ it.  Others
>>                 
> could
>               
>> choose not to use it.
>>
>>
>>
>> Chuck
>>
>>
>>
>> From: tdwg-guid-bounces@lists.tdwg.org [mailto:tdwg-guid- 
>> bounces@lists.tdwg.org] On Behalf Of Ricardo Pereira
>> Sent: Friday, July 13, 2007 9:47 AM
>> To: tdwg-guid@lists.tdwg.org
>> Subject: [tdwg-guid] LSID metadata persistence (or 
>>
lack thereof)
...
...
...
...
>>
>>     Hi there folks,
>>
>>     As Chuck mentioned a few weeks ago, we do have a few 
>> outstanding issues to address regarding LSIDs. I 
>>
would like to
...
...
...
...
>> discuss those one by one, in an orderly manner, and reach
>>                 
> consensus
>               
>> as much as we can. Then we can sum them up in a TDWG 
>>
standard,
...
...
...
...
>> possibly by or shortly after the Bratislava conference.
>>
>>     The first issue I would like to discuss is LSID metadata 
>> persistence. First, let me remind you of a corollary
>>                 
> established by
>               
>> the LSID specification:
>>
>>             Corollary 1: LSIDs are not guaranteed to be
>>                 
> resolvable
>               
>> indefinitely.
>>
>>     In other words, there is no guarantee that one will
>>                 
> always be
>               
>> able to retrieve the data associated with an LSID as the
>>                 
> authority
>               
>> may choose (or be forced) not  to resolve an LSID anymore.
>>
>>     Second, let me distinguish this kind of persistence I'm
>>                 
> talking
>               
>> about from other two related concepts (which we'll not
>>                 
> discuss in
>               
>> this thread):
>>
>>         1) Persistence of Assignment: Once assigned to an
>>                 
> object,
>               
>> an LSID is indefinitely associated with it. The same LSID
>>                 
> cannot be
>               
>> assigned to another object. Ever! The LSID may not be 
>>
resolvable
...
...
...
...
>> anymore, but it cannot be assigned to another object. This is 
>> established by the LSID specification.
>>
>>         2) Persistence of LSID Data: The data 
>>
associated with an
...
...
...
...
>> LSID (i.e, the byte stream returned by the LSID getData call)
>>                 
> must
>               
>> never change. Although the LSID may not be resolvable anymore 
>> (according to corollary 1), the data associated with an LSID
>>                 
> must
>               
>> never ever change. That's defined by the LSID spec, too.
>>
>>     What I want to discuss here is the persistence of LSID
>>                 
> metadata
>               
>> (what is returned by the getMetadata call) or the 
>>
lack thereof.
...
...
...
...
>>     A use case associated with metadata persistence is when
>>                 
> someone
>               
>> collects observation records (and implicitly, their
>>                 
> determinations)
>               
>> and runs an experiment (a model or simulation) with it. This
>>                 
> person
>               
>> may want to record the identifiers of the points used so that 
>> someone using the results of that experiment may refer back
>>                 
> to the
>               
>> primary data, to validate or repeat it the experiment.
>>
>>     The bad news is that LSID identification scheme (or any
>>                 
> other
>               
>> GUID that I know of) was not designed to guarantee metadata 
>> persistence, and thus it cannot implement the use 
>>
case above by
...
...
...
...
>> itself. To implement that use case, the specification would
>>                 
> have to
>               
>> guarantee that the metadata (which we are using here 
>>
as data) is
...
...
...
...
>> immutable. But it doesn't.
>>
>>     Most of us wish that metadata was persistent, but 
>>
it isn't.
...
...
...
...
>> Many things can change in the metadata: a new 
>>
determination, a
...
...
...
...
>> mispeling that is corrected, many things. We just cannot
>>                 
> guarantee
>               
>> that the metadata will look like it was sometime ago.
>>
>>     We then reach the following conclusion.
>>
>>             Corollary 2: LSIDs metadata is not immutable nor 
>> persistent.
>>
>>     The consequence of this corollary is that, if you need to
>>                 
> refer
>               
>> back to a piece of information (metadata) associated with an
>>                 
> LSID,
>               
>> exactly as it was when you got it, you must make a copy of
>>                 
> it, or
>               
>> arrange that someone else make that copy for you.
>>
>>     In other words, a client cannot assume that the metadata 
>> associated with an LSID today will be the same 
>>
tomorrow. If the
...
...
...
...
>> client does assume that, it may be relying on a false 
>>
assumption
...
...
...
...
>> and its output may be flawed.
>>
>>     If we are not happy with that conclusion, we may 
>>
develop an
...
...
...
...
>> additional component in our architecture, an archive of some
>>                 
> sort,
>               
>> to handle (meta)data persistence. That is exactly what the
>>                 
> STD-DOI
>               
>> project (http://www.std-doi.de/) and SEEK (http://
>> seek.ecoinformatics.org) have done to some extent.
>>
>>     While we cannot guarantee that LSID metadata is
>>                 
> persistent nor
>               
>> immutable, we can definitely document how the metadata have
>>                 
> changed
>               
>> through metadata versioning. That's the topic of the next
>>                 
> thread.
>               
>> We will move on to discuss metadata versioning as 
>>
soon as we are
...
...
...
...
>> done with metadata persistence.
>>
>>     Cheers,
>>
>> Ricardo
>>
>> _______________________________________________
>> tdwg-guid mailing list
>> tdwg-guid@lists.tdwg.org
>> http://lists.tdwg.org/mailman/listinfo/tdwg-guid
>>                 
> _______________________________________________
> tdwg-guid mailing list
> tdwg-guid@lists.tdwg.org
> http://lists.tdwg.org/mailman/listinfo/tdwg-guid
>
>
> P Think Green - don't print this email unless you really need to
>
>
>

...
...
...
...
> ******
>  The information contained in this e-mail and any files 
> transmitted with it is confidential and is for the 
>
exclusive use
...
...
...
...
> of the intended recipient. If you are not the intended 
>
recipient
...
...
...
...
> please note that any distribution, copying or use of this 
> communication or the information in it is prohibited.
>
>  Whilst CAB International trading as CABI takes steps 
>
to prevent
...
...
...
...
> the transmission of viruses via e-mail, we cannot 
>
guarantee that
...
...
...
...
> any e-mail or attachment is free from computer viruses 
>
and you are
...
...
...
...
> strongly advised to undertake your own anti-virus precautions.
>
>  If you have received this communication in error, 
>
please notify
...
...
...
...
> us by e-mail at cabi@cabi.org or by telephone on +44 (0)1491 
> 829199 and then delete the e-mail and any copies of it.
>
>  CABI is an International Organization recognised by the UK 
> Government under Statutory Instrument 1982 No. 1071.
>
>
>

...
...
...
...
> ********
> _______________________________________________
> tdwg-guid mailing list
> tdwg-guid@lists.tdwg.org
> http://lists.tdwg.org/mailman/listinfo/tdwg-guid
>
>
>               
--Robert A. Morris
Professor of Computer Science
UMASS-Boston
ram@cs.umb.edu
http://bdei.cs.umb.edu/
http://www.cs.umb.edu/~ram
http://www.cs.umb.edu/~ram/calendar.html
phone (+1)617 287 6466

tdwg-guid mailing list
tdwg-guid@lists.tdwg.org
http://lists.tdwg.org/mailman/listinfo/tdwg-guid

tdwg-guid mailing list
tdwg-guid@lists.tdwg.org
http://lists.tdwg.org/mailman/listinfo/tdwg-guid

tdwg-guid mailing list
tdwg-guid@lists.tdwg.org
http://lists.tdwg.org/mailman/listinfo/tdwg-guid

tdwg-guid mailing list
tdwg-guid@lists.tdwg.org
http://lists.tdwg.org/mailman/listinfo/tdwg-guid