Issues related to updating data in native XML database systems are studied extensively nowadays. In this work we consider a problem of updating typed XML documents having their schema described by a Document Type Definition (DTD) without breaking their validity and with ensured transaction consistency. We present a way how to express constructs available in DTD by using a functional framework and propose algorithms for performing insert, replace and delete operations. This solution is an intermediate step we need for our ongoing research - formal comparison of XQuery and XML-λ. ∗ I would like to thank to Prof. Pokorny´ for his patience and provisioning of many helpful hints for my research.
Fundamental work we continue to work on is Pokorny´ ’s
proposal of a functional framework for modeling and
querying XML – XML-λ [
This work directly continues in the topic that we have
opened in [
Nevertheless, our primary motivation is not to develop a totally new sort of an XML update language but rather to propose an update extension that allows us to go on with our planned research in the future – comparison of properties of XQuery and XML-λ and evaluation of potential mutual transformations of queries written in respective languages. We see the benefit of this paper par2
By the term updating XML we mean the ability of a language to perform modifications (insert, replace and delete operations) over an XML document or a collection of XML documents.
Since the creation of the XML in 1998, there have
been many efforts to develop various data models and
query languages for databases of XML data. Multiple
approaches for indexing and query optimizations have
been invented. On the other hand, the problem of
updating XML gains more interest in few past years. Yet
there seems to be not a complete solution for this
problem. Existing papers dealing with updating XML are
mostly related to XQuery [
For the sake of completeness we should not omit
XUpdate [
Another research field is represented by XDuce [
Considering previous works we can deduce that there are common types of operations for performing modifications that are to be embedded in a language – delete, replace, insert-before, insert-after or insert-as-child. This seems to be a sufficient base for ongoing work. None of those proposals but deals in detail with the problem of updating typed data and hence it makes sense to put effort and study this problem. 3
XML-λ is a proposal published by Pokorny´ [
Except of the original proposal, that defines its formal
base and shows its usage primarily as a query language
for XML, there is a consecutive work that introduces
updates into the language available in [
Here, we focus primarily on extending and improving the update part of the framework. Basic facts about the framework are repeated in following sections rather for convenience. 3.1
In XML-λ there are three important components related to its type system: element types, element objects and elements. We can imagine these components as the data dictionary in relational database systems. Note also Figure 1 for relationships of basic terms between W3C standards and the XML-λ Framework.
Element types are derived from a particular DTD and in our scenario they cannot be changed – we do not allow any schema changes but only data modifications. For each element defined in the DTD there exists exactly one element type in the set of all available element types (called TE).
Consequently, we denote E as a set of abstract elements. Set members are of element types. Note that (from definition) E is an infinite set.
Element objects1 are basically functions of type either E → String or E → (E × . . . × E). Application of 1We denote the element object of type t ∈ TE as t-object these functions to an abstract element allows access to element’s content. Elements are, informally, values of element objects, i.e. of functions. For each t ∈ TE there exists a corresponding t-object.
For convenience, we add a ”nullary function” (also known as 0-ary function) into our model. This function returns a set of all abstract elements of a given element type from an XML document.
Finally, we can say that in XML-λ the instance of an XML document is represented by a subset of E and set of respective t-objects.
For readers familiar with W3C terminology, there is a comparison of related terms in both environments shown in Table 1. 3.2
This section shows an example of using the XML-λ Framework in a real example with detailed description. Let us consider an example DTD shown in Figure 2. <!ELEMENT bib <!ELEMENT book <!ELEMENT author <!ELEMENT title <!ELEMENT last <!ELEMENT first <!ELEMENT price (book* )> (title, author+, price )>
(last, first )> (#PCDATA )> (#PCDATA )> (#PCDATA )> (#PCDATA )>
For given schema we obtain element types as follows: BIB : BOOK∗, BOOK : (T IT LE, AU T HOR+, P RICE), AU T HOR : (LAST , F IRST ), LAST : String, F IRST : String, T IT LE : String, P RICE : String.
Then, we define functional types – designated as t-objects: BIB : E → 2E, BOOK : E → (E × 2E × E), AU T HOR : E → (E × E), T IT LE : E → String, LAST : E → String, F IRST : E → String, P RICE : E → String.
These types are the cornerstone for manipulation with typed data from XML documents as shown in the list of semantic functions (see Table 2).
Having look at DTD in Figure 2 and sample data in Figure 3 we can obviously see that there are 7 abstract elements (members of E0 ⊂ E). Now, for instance, the title-object is defined exactly for one abstract element (the one gained from <title>TCP/IP Illustrated</title> element and for this abstract element it returns a string value ”TCP/IP Illustrated”.
XML 1.0 Document Type Definition (DTD) DOM - A tree instance XPath, XQuery, XSLT XML 1.0 Types in TE derived from DTD Set of abstract elements – E, definition of t-objects Simply typed lambda calculus <bib> <book> <title>TCP/IP Illustrated</title> <author> <last>Stevens</last> <first>W.</first> </author> <price>65.95</price> </book> ... </bib>
Following example query returns all books with
specified price
lambda b (/book(b) and b/price = "65.95" )
Evaluation of this query with respect to semantics
described in [
For readers familiar with XQuery, here is the same query expressed in XQuery syntax: { } for $b in doc("bib.xml")/bib/book where $b/price = "65.95" return {$b}
Expected output is shown in Figure 4. 4
This section covers the process of updating data in an existing XML data store. Thus, we do not update XML schema of these documents but their content only. It is a typical database life cycle – the database schema remains <book> <title>TCP/IP Illustrated</title> <author> <last>Stevens</last> <first>W.</first> </author> <price>65.95</price> </book> stable but the data is changing in time. In our (query and update) language there is no way how to construct new documents yet – sometimes this approach is called ”incremental update”. In other words we can change the structure of the input document (w.r.t the DTD) by a given XML-λ update statement but cannot e.g. create a set of new XML files. We can describe the whole operation of updating an XML document rather on a physical level as (1) retrieving its content from database, (2) performing update, (3) storing document back to database. This paper deals with the second part of the process. Viewed from closer look in more detailed pieces it is (a) localization of point in data model where the change will take place, (b) validation of requested operation, (c) execution of the update operation. These steps are shown more from the semantical point of view, in implementation it is usually not necessary to retrieve complete XML document from database into memory but we can manipulate only with a part of its content needful for update.
There are two options when to perform data validation – before or after an update. The XQuery Update Facility proposal uses optional post-update revalidation; in our approach we focus more on doing pre-update checks. Our goal is to detect the maximum number of possible conflicts during compilation of the update statement and potentially raise a static error. Unfortunately, not in all cases is the information from data model enough for validation and, therefore, it is necessary to perform validation with respect to particular data stored in the data store. We discuss this issue later in Section 5. Regardless the scenario, the processed XML document is a valid instance in the type system both before and after update. 1. Elements constraints - Specify the type of element content. The possible value is one of EMPTY, ANY, MIXED or ELEMENT_CONTENT, 2. Structure constraints - The occurrence of elements in a content model. Options are exactly-one, zeroor-one, zero-or-more, one-or-more, 3. Attributes constraints - Each attribute can have one of #REQUIRED, #IMPLIED, #FIXED, ID, IDREF(S) options assigned.
Each update operation can or cannot be affected by any construct from the particular DTD. Note that element content type ANY cannot be used in XML-λ, because of the framework’s type system nature. 4.3
This section covers the basic concept of updates in the
XML-λ Framework. It initially had not have any update
facility. We had to extend it with features allowing us to
check constraints available in DTD. The idea of updates
has been opened in [
As already outlined in Section 3.1, there are three crucial components related to the type system - element types, element objects and abstract elements. Element types are derived from a particular DTD and in our scenario they cannot be changed.
Elements are, informally, values of element objects, i.e. of functions. Thus, by updating an XML document in XML-λ we modify the actual domains of these functions (subsets of E) and element objects affected by required update operation (insert, delete, replace).
Before of that, we have to validate requested operation. For now let us consider constraints described by a DTD but in the outlook there are more options which standards we plan to use as well (e.g. XML Schema). Therefore we design our solution keeping this possibility in mind.
Sections 5.3 - 5.5 discuss the semantics of delete, insert and replace operations in detail. 4.4
One disadvantage of the solution proposed in [
For now, we can consider that a transaction manager locks the complete part of XML data that can be modified during the update operation (in the worst case even the whole XML document). It is a significant performance issue but for purpose of this paper it is not fundamental.
Thus, at the beginning of suggested algorithms we only ask for locking of a specific part of processed XML document and keep all concurency-related worries and issues on the ”virtual” transaction manager. 5
XML-λ
and then semantics of all supported update operations – insert, delete and replace. In this paper we base our work on constraints available in DTDs. The goal here is to describe these limitations in general as much as possible for eventual future extensions. Validating update operations is a problem very closely related to the problem of validating a complete XML instance. This process, however, can be for extensive documents very time consuming.
In our approach we propose two sets of types and algorithms for validation for each update operation. Mentioned sets are constructed and initiated during analysis of given DTD and contain element types from TE . Due to the fact that we do not allow schema changes, they are stable in time.
1. Timmutable. Abstract elements of types from this set and respective t-objects are not changeable in our data model. In terms of DTD these types are associated with DTD types which content cannot be modified, i.e. attributes declared as #FIXED and element types with EMPTY content model. 2. Tmandatory. Abstract elements of types from this set and respective t-objects are not modifiable (must not be removed) in our data model. In terms of DTD this set contains types associated with attribute types with #REQUIRED declaration and element types for those types Ti iff all occurrences of Ti in given DTD are exactly-one.
These sets we use in our semantics for particular update operations. For future work we can also consider sets Treferencing and Treferenced of types associated with attributes declared in given DTD as IDREF or IDREFS and for attributes declared as ID respectively. In following text we use number of functions with informal meaning as summarized in Table 2. 5.2
General Notes to Proposed Algorithms Following sections contain particular algorithms for data modification shown in detail. The most important of them – Delete and Insert – follow the same structure. First, they check (optionally) the validity of the operation and if there is no conflict with the type system definition they break down the modification into a list of primitive operations (stored in a structure also known as the ”Pending Update List”). This list represents hence the result of these algorithms. Note that the Replace algorithm combines aforesaid Delete and Insert with within. The items inside the list are pairs (e, op), where e ∈ E is an abstract element and op ∈ {DELET E, IN SERT } is the operation to be carried out.
The output pending update list, that represents the result of each update algorithm, is then passed to the ProcessPendingList algorithm. This algorithm then executes all primitive changes requested. 5.3
In this section we describe two parts of the update process – general concept of validation we use in XML-λ Formally, we decompose the process into two parts – a function checkDelete that is used for checking whether parent(e) typeOf(e) cardMin(e),cardMax(e) alterTObjectDel(e, t), alterTObjectIns(e, t) isSubtype(t1, t2) canSubstitute(t1, t2) isElementary(t) application(e, t) projection(n-tuple, t) count(n-tuple) For an e ∈ E returns its parent abstract element. An abstract element can have at most one associated ”parent” element. When considering E as infinite set of abstract elements, most of them have no parent associated.
For an e ∈ E returns its element type (see Section 3.1).
Return minimal (or maximal, respectively) cardinality of an abstract element’s type in a particular data model instance.
Alters the t-object for given e ∈ E. Regarding the fact that t-objects are functions these semantic functions change the domain of given t-object and thus associations among abstract elements. Basically, alterTObjectDel removes the abstract element e from domain of the t-object and alterTObjectIns adds the abstract element e into the domain.
Describes a relation between element types t1 and t2. Returns true iff the result of application(e, t2) for an e ∈ E can return an n-tuple containing an abstract element of type t1 (at any position).
Returns true iff an abstract element e1 of type t1 can replace an element e2 of type t2 without breaking document’s validity. It is utilized in the Replace algorithm.
Returns true iff t is an elementary element type.
Executes an application of t-object to the e element. In general it returns an n-tuple from Cartesian product of (E × . . . × E).
Note that the application function serves for diving in the ”content” of an element. Retrieves all elements of type t from given n-tuple.
Returns number of elements in an n-tuple. an abstract element can be deleted and a complete algorithm Delete that accomplishes the operation completely: Function: checkDelete; Input: E - set of abstract elements
e - an abstract element to be deleted Output: returns true - deletion is allowed,
false - deletion is denied begin let t = typeOf (e); if ((t ∈ Tmandatory) or (t ∈ Timmutable)) then
return false; if ((cardM in(e) = 0) and (cardM ax(e) = ∞)) then
return true; if ((cardM in(e) ≥ 1) and (count(application(parent(e), t)) > 1)) then return false; return true; end
Input: E - set of abstract elements e - an abstract element to be deleted checkV alidity - a boolean flag. Enables or disables validity check. Default is true. trans - a new transaction pList - a list of currently pending update
operations Output: returns true - delete is allowed,
false - delete failed pList - updated list of pending operations begin /* Lock the data being deleted */ trans.lockRequest(DELET E N ODE, e); /* Check type constraints - if requested */ if (checkV alidity) then if (not checkDelete(E, e)) then return false; let S = new Stack(); S.push(e); while (tmp = S.pop()) do let t = typeOf (tmp); let nt = application(tmp, t); For i = 1 to count(nt) let etmp = nt[i]; let ttmp = typeOf (etmp); /* Elementary element types are added into the pending delete list */ if (isElementary(ttmp)) then
pList.add(etmp, DELET E)) else /* Complex element types are stored for
next iterations */
S.push(etmp); next; end /* Add the initial abstract element to pending list */ pList.add(tmp, DELET E); /* Deletion is finished */ return true; end 5.4
As for the Delete algorithm, we propose two parts of the insert process – function checkInsert that validates insertion of given abstract element and Insert algorithm that implements the operation in whole.
Function: checkInsert; Input: E - set of abstract elements e1 - an abstract element to be inserted, e2 - an abstract element to be associated with e1 as its parent abstract element, Output: returns true - insertion is allowed,
false - insertion is denied begin let t = typeOf (e2); if (t ∈ Timmutable) then return false; /* Add the initial abstract element to pending list */ pList.add(e1, IN SERT ) /* Insert is finished */ return true; end /*Traversing through all ”sibling” abstract elements*/ let nt = application(tmp, t); for i = 1 to count(nt) let etmp = nt[i]; let ttmp = typeOf (etmp); if (isSubtype(typeOf (e1), ttmp)) then if (cardM ax(etmp) > 1) then return true; if ((cardM in(e) = 0) and (cardM ax(etmp = ∞)) and /* Lock the data being inserted */ trans.lockRequest(IN SERT N ODE, e1); if (checkV alidity) then
if not checkInsert(E, e1, e2) then return false; let S = new Stack(); S.push(e); while (tmp = S.pop()) do let t = typeOf (tmp); let nt = application(tmp, t); For i = 1 to count(nt) let etmp = nt[i]; let ttmp = typeOf (etmp); /* Elementary element types are inserted
into pending list */ if (isElementary(ttmp)) then
pList.add(etmp, IN SERT ) else /* Complex element types are stored for
next iterations */
S.push(etmp); next; end type tnew (for example, for t1 = (a|b), t2 = a ⇒ canSubstitute(t1, t2) = true).
Note that we turn off the type validation for particular Delete and Insert calls. Type validity is already checked at the beginning of the algorithm.
Input: E - set of abstract elements e1 - an abstract element to be replaced, e2 - an abstract element used as the substitution of e1 trans - a new transaction, pList - a list of currently pending update operations Output: returns true - replace is allowed,
false - replace failed pList - list of pending update operations begin /* It must be allowed to replace e1 with e2 */ if not (canSubstitute(typeOf (e1), typeOf (e2)) then return false; let etmp = parent(e1); if not Delete(E, e1, trans, f alse, pList) then return false; if not Insert(E, e2, etmp, trans, f alse, pList) then return false; end /* Replace is finished */ return true; 5.6
The Delete, Insert and Replace algorithms introduced in previous sections transform high-level manipulation operations into a sequential list of primitives that is stored in the structure called Pending Update List – here it is denoted as variable pList. This list is to be processed by the database engine at the end of each high-level operation in cooperation with the transaction manager.
Following algorithm describes the operation more formally.
Algorithm: ProcessPendingList; Input: E - set of abstract elements t-objects associated with affected abstract elements pList - a list of currently pending update operations Output: pList - an empty pending list, begin
E - (potentially modified) set of abstract elements, t-objects - (potentially modified) t-objects while (pList.hasN ext()) do let tmp = pList.next(); pList.remove(); let e = tmp.getItem(); let t = typeOf (parent(e)); let op = tmp.getOperation(); if (op == IN SERT ) then let E = E ∪ e; alterT ObjectIns(e, t); else / ∗ DELET E ∗ / let E = E \ e; alterT ObjectDel(e, t); next; end /* Pending List is now empty */ end 5.7
Considering the XML-λ Query Language as specified
in [
By the extensions proposed in this paper we obtain a framework suitable for both querying and updating XML data. With respect to its original idea there is a number of potential applications of the framework. Let us sketch three possible ways how to continue with its development: 1. further expand its query and update capabilities, 2. use it for integration of heterogeneous data sources, 3. use the XML-λ’s formal apparatus for description of XQuery semantics.
For each option there is still a lot of work ahead. To get a complete query framework we have to finalize an issue with references within documents (IDs and IDREFS). This is only a technical problem of introducing new types and formalizing the algorithm to be executed to keep the documents consistent and valid. Let us also note another questionable area that is not covered in this paper and thus the dependencies of multiple update operations in one ”query” statement. This issue deals with transactional processing and optimizing multiple update primitives’ execution.
This option is also questionable because of wide acceptance of XQuery as the de-facto theoretical and industrial standard in the area of query languages for XML. At least, the research here will require extensive enthusiasm and sufficient resources.
Integration of heterogeneous data sources (as outlined
in [
The third option for ongoing research is using the
framework for description of XQuery’s semantics. This
is probably the most interesting research branch from the
theoretical point of view. It generally means that we will
be able to express any XQuery statement with a
corresponding XML-λ alternative. This idea represents a
theoretical research related to formal methods and
compilers. In this case the framework is going to be used as a
tool for transformation of queries between various query
and update languages. In addition, we can use this tool
for evaluation of XQuery queries within our prototype
of a native XML database system ExDB [
Another feasible challenge for future work is redefinition of the type system by replacement of DTD types by types available in XML Schema or in RELAX NG. This means restructuralization of the type systems Treg and TE and redevelopment of the idea of constraint sets. This research would demonstrate that the concept of functional framework is not strictly bound to DTD but is more general as we declare. 7
We have shown a proposal for updating typed XML data constrained by a Document Type Definition. We build on a functional framework for querying XML that can utilize concept of DTD constraints in its type system TE. Main part of the paper discusses the idea of extending the framework with update operations with accent to keep the documents always valid. By enriching the XML-λ query language with modification operations - inserts, deletes and replacements - we obtain a language suitable both for querying and updating XML documents.
Further work and research directions outlined in Section 6 lead to onward framework extensions – either improving its query capabilities, using it for integration of heterogeneous data or utilizing the framework for description of semantics of various query languages.
In any case the work presented in this paper creates sufficient base for extensive future work.