4 General Notes for Implementers

This section discusses a number of topics other than the exchange of data sets in SDMX formats. Supported only in SDMX-ML (and some in SDMX-JSON), these topics include the use of the reference metadata mechanism in SDMX, the use of Structure Sets and Reporting Taxonomies, the use of Processes, a discussion of time and datatyping, and the conventional mechanisms within the SDMX-ML Structure message regarding versioning and referencing.

4.1 Representations

This section does not go into great detail on these topics but provides a useful overview of these features to assist implementors in further use of the parts of the specification which are relevant to them.

There are several different representations in SDMX-ML, taken from XML Schemas and common programming languages. The table below describes the various representations, which are found in SDMX-ML, and their equivalents.

There are also a number of SDMX-ML data types which do not have these direct correspondences, often because they are composite representations or restrictions of a broader data type. For most of these, there are simple types which can be referenced from the SDMX schemas, for others a derived simple type will be necessary:

• AlphaNumeric (common:AlphaNumericType, string which only allows A-z and 09)
• Alpha (common:AlphaType, string which only allows A-z)
• Numeric (common:NumericType, string which only allows 0-9, but is not numeric so that is can having leading zeros)
• Count (xs:integer, a sequence with an interval of "1")
• InclusiveValueRange (xs:decimal with the minValue and maxValue facets supplying the bounds)
• ExclusiveValueRange (xs:decimal with the minValue and maxValue facets supplying the bounds)
• Incremental (xs:decimal with a specified interval; the interval is typically enforced outside of the XML validation)
• TimeRange (common:TimeRangeType, startDateTime + Duration)
• ObservationalTimePeriod (common:ObservationalTimePeriodType, a union of StandardTimePeriod and TimeRange).
• StandardTimePeriod (common:StandardTimePeriodType, a union of BasicTimePeriod and ReportingTimePeriod).
• BasicTimePeriod (common:BasicTimePeriodType, a union of GregorianTimePeriod and DateTime)
• GregorianTimePeriod (common:GregorianTimePeriodType, a union of GregorianYear, GregorianMonth, and GregorianDay)
• ReportingTimePeriod (common:ReportingTimePeriodType, a union of ReportingYear, ReportingSemester, ReportingTrimester, ReportingQuarter, ReportingMonth, ReportingWeek, and ReportingDay).
• ReportingYear (common:ReportingYearType)
• ReportingSemester (common:ReportingSemesterType)
• ReportingTrimester (common:ReportingTrimesterType)
• ReportingQuarter (common:ReportingQuarterType)
• ReportingMonth (common:ReportingMonthType)
• ReportingWeek (common:ReportingWeekType)
• ReportingDay (common:ReportingDayType)
• XHTML (common:StructuredText, allows for multi-lingual text content that has XHTML markup)
• KeyValues (common:DataKeyType)
• IdentifiableReference (types for each IdentifiableObject)
• GeospatialInformation (a geo feature set, according to the pattern in section 7.2)

Data types also have a set of facets:

• isSequence = true | false (indicates a sequentially increasing value)
• minLength = positive integer (# of characters/digits)
• maxLength = positive integer (# of characters/digits)
• startValue = decimal (for numeric sequence)
• endValue = decimal (for numeric sequence)
• interval = decimal (for numeric sequence) • timeInterval = duration
• startTime = BasicTimePeriod (for time range)
• endTime = BasicTimePeriod (for time range)
• minValue = decimal (for numeric range)
• maxValue = decimal (for numeric range)
• decimal = Integer (# of digits to right of decimal point)
• pattern = (a regular expression, as per W3C XML Schema)
• isMultiLingual = boolean (for specifying text can occur in more than one language)

Note that code lists may also have textual representations assigned to them, in addition to their enumeration of codes.

4.1.1 Data Types

XML and JSON schemas support a variety of data types that, although rich, are not mapped one-to-one in all cases. This section provides an explanation of the mapping performed in SDMX 3.0, between such cases.

For identifiers, text fields and Codes there are no restriction from either side, since a generic type (e.g., that of string) accompanied by the proper regular expression works equally well for both XML and JSON.

For example, for the id type, this is the XML schema definition:

 <xs:simpleType name="IDType">
 <xs:restriction base="NestedIDType">
 <xs:pattern value="[A-Za-z0-9_@$\-]+"/>
 </xs:restriction>
 </xs:simpleType>

Where the NestedIDType is also a restriction of string.

The above looks like this, in JSON schema:

 "idType": {
    "type": "string",
    "pattern": "^[A-Za-z0-9_@$-]+$"
 }

There are also cases, though, that data types cannot be mapped like above. One such case is the array data type, which was introduced in SDMX 3.0 as a new representation. In JSON schema an array is already natively foreseen, while in the XML schema, this has to be defined as a complex type, with an SDMX specific definition (i.e., specific element/attribute names for SDMX). Beyond that, the minimum and/or maximum number of items within an array is possible in both cases.

Further to the above, the mapping between the non-native data types is presented in the table below:

4.2 Time and Time Format

This section does not go into great detail on these topics but provides a useful overview of these features to assist implementors in further use of the parts of the specification which are relevant to them.

4.2.1 Introduction

First, it is important to recognize that most observation times are a period. SDMX specifies precisely how Time is handled.

The representation of time is broken into a hierarchical collection of representations. A data structure definition can use of any of the representations in the hierarchy as the representation of time. This allows for the time dimension of a particular data structure definition allow for only a subset of the default representation.

The hierarchy of time formats is as follows (bold indicates a category which is made up of multiple formats, italic indicates a distinct format):

• Observational Time Period
  • Standard Time Period
    • Basic Time Period
      • Gregorian Time Period
      • Date Time
    • Reporting Time Period
  • Time Range

The details of these time period categories and of the distinct formats which make them up are detailed in the sections to follow.

4.2.2 Observational Time Period

This is the superset of all time representations in SDMX. This allows for time to be expressed as any of the allowable formats.

4.2.3 Standard Time Period

This is the superset of any predefined time period or a distinct point in time. A time period consists of a distinct start and end point. If the start and end of a period are expressed as date instead of a complete date time, then it is implied that the start of the period is the beginning of the start day (i.e. 00:00:00) and the end of the period is the end of the end day (i.e. 23:59:59).

4.2.4 Gregorian Time Period

A Gregorian time period is always represented by a Gregorian year, year-month, or day. These are all based on ISO 8601 dates. The representation in SDMX-ML messages and the period covered by each of the Gregorian time periods are as follows:

Gregorian Year:
Representation: xs:gYear (YYYY)
Period: the start of January 1 to the end of December 31
Gregorian Year Month:
Representation: xs:gYearMonth (YYYY-MM)
Period: the start of the first day of the month to end of the last day of the month
Gregorian Day:
Representation: xs:date (YYYY-MM-DD)
Period: the start of the day (00:00:00) to the end of the day (23:59:59)

4.2.5 Date Time

This is used to unambiguously state that a date-time represents an observation at a single point in time. Therefore, if one wants to use SDMX for data which is measured at a distinct point in time rather than being reported over a period, the date-time representation can be used.

Representation: xs:dateTime (YYYY-MM-DDThh:mm:ss)²

4.2.6 Standard Reporting Period

Standard reporting periods are periods of time in relation to a reporting year. Each of these standard reporting periods has a duration (based on the ISO 8601 definition) associated with it. The general format of a reporting period is as follows:

[REPORTING_YEAR]-[PERIOD_INDICATOR][PERIOD_VALUE]

Where:

REPORTING_YEAR represents the reporting year as four digits (YYYY) PERIOD_INDICATOR identifies the type of period which determines the duration of the period
PERIOD_VALUE indicates the actual period within the year

The following section details each of the standard reporting periods defined in SDMX:

Reporting Year:
Period Indicator: A
Period Duration: P1Y (one year)
Limit per year: 1
Representation: common:ReportingYearType (YYYY-A1, e.g. 2000-A1)
Reporting Semester:
Period Indicator: S
Period Duration: P6M (six months)
Limit per year: 2
Representation: common:ReportingSemesterType (YYYY-Ss, e.g. 2000-S2)
Reporting Trimester:
Period Indicator: T
Period Duration: P4M (four months)
Limit per year: 3
Representation: common:ReportingTrimesterType (YYYY-Tt, e.g. 2000-T3)
Reporting Quarter:
Period Indicator: Q
Period Duration: P3M (three months)
Limit per year: 4
Representation: common:ReportingQuarterType (YYYY-Qq, e.g. 2000-Q4)
Reporting Month:
Period Indicator: M
Period Duration: P1M (one month)
Limit per year: 1
Representation: common:ReportingMonthType (YYYY-Mmm, e.g. 2000-M12) Notes: The reporting month is always represented as two digits, therefore 1-9 are 0 padded (e.g. 01). This allows the values to be sorted chronologically using textual sorting methods.
Reporting Week:
Period Indicator: W
Period Duration: P7D (seven days)
Limit per year: 53
Representation: common:ReportingWeekType (YYYY-Www, e.g. 2000-W53)
Notes: There are either 52 or 53 weeks in a reporting year. This is based on the ISO 8601 definition of a week (Monday - Saturday), where the first week of a reporting year is defined as the week with the first Thursday on or after the reporting year start day.³ The reporting week is always represented as two digits, therefore 1-9 are 0 padded (e.g. 01). This allows the values to be sorted chronologically using textual sorting methods.
Reporting Day:
Period Indicator: D
Period Duration: P1D (one day)
Limit per year: 366
Representation: common:ReportingDayType (YYYY-Dddd, e.g. 2000-D366) Notes: There are either 365 or 366 days in a reporting year, depending on whether the reporting year includes leap day (February 29). The reporting day is always represented as three digits, therefore 1-99 are 0 padded (e.g. 001). This allows the values to be sorted chronologically using textual sorting methods.
The meaning of a reporting year is always based on the start day of the year and requires that the reporting year is expressed as the year at the start of the period. This start day is always the same for a reporting year, and is expressed as a day and a month (e.g. July 1). Therefore, the reporting year 2000 with a start day of July 1 begins on July 1, 2000.

A specialized attribute (reporting year start day) exists for the purpose of communicating the reporting year start day. This attribute has a fixed identifier (REPORTING_YEAR_START_DAY) and a fixed representation (xs:gMonthDay) so that it can always be easily identified and processed in a data message. Although this attribute exists in specialized sub-class, it functions the same as any other attribute outside of its identification and representation. It must takes its identity from a concept and state its relationship with other components of the data structure definition. The ability to state this relationship allows this reporting year start day attribute to exist at the appropriate levels of a data message. In the absence of this attribute, the reporting year start date is assumed to be January 1; therefore if the reporting year coincides with the calendar year, this Attribute is not necessary.

Since the duration and the reporting year start day are known for any reporting period, it is possible to relate any reporting period to a distinct calendar period. The actual Gregorian calendar period covered by the reporting period can be computed as follows (based on the standard format of [REPROTING_YEAR]-[PERIOD_INDICATOR][PERIOD_VALUE] and the reporting year start day as [REPORTING_YEAR_START_DAY]):

1. Determine [REPORTING_YEAR_BASE]:
Combine [REPORTING_YEAR] of the reporting period value (YYYY) with [REPORTING_YEAR_START_DAY] (MM-DD) to get a date (YYYY-MM-DD).
This is the [REPORTING_YEAR_START_DATE]

a) If the [PERIOD_INDICATOR] is W:

1. If [REPORTING_YEAR_START_DATE] is a Friday, Saturday, or Sunday:
Add^[4] (P3D, P2D, or P1D respectively) to the 
[REPORTING_YEAR_START_DATE]. The result is the [REPORTING_YEAR_BASE].

2. If [REPORTING_YEAR_START_DATE] is a Monday, Tuesday, Wednesday, or Thursday:

Add⁴ (P0D, -P1D, -P2D, or -P3D respectively) to the
[REPORTING_YEAR_START_DATE]. The result is the [REPORTING_YEAR_BASE].

b) Else:

The [REPORTING_YEAR_START_DATE] is the [REPORTING_YEAR_BASE].

2. Determine [PERIOD_DURATION]:

1. If the [PERIOD_INDICATOR] is A, the [PERIOD_DURATION] is P1Y.
2. If the [PERIOD_INDICATOR] is S, the [PERIOD_DURATION] is P6M.
3. If the [PERIOD_INDICATOR] is T, the [PERIOD_DURATION] is P4M.
4. If the [PERIOD_INDICATOR] is Q, the [PERIOD_DURATION] is P3M.
5. If the [PERIOD_INDICATOR] is M, the [PERIOD_DURATION] is P1M.
6. If the [PERIOD_INDICATOR] is W, the [PERIOD_DURATION] is P7D.
7. If the [PERIOD_INDICATOR] is D, the [PERIOD_DURATION] is P1D.
8. Determine [PERIOD_START]:

Subtract one from the [PERIOD_VALUE] and multiply this by the [PERIOD_DURATION]. Add⁴ this to the [REPORTING_YEAR_BASE]. The result is the [PERIOD_START].

1. Determine the [PERIOD_END]:

Multiply the [PERIOD_VALUE] by the [PERIOD_DURATION]. Add⁴ this to the [REPORTING_YEAR_BASE] add⁴ -P1D. The result is the [PERIOD_END].

For all of these ranges, the bounds include the beginning of the [PERIOD_START] (i.e. 00:00:00) and the end of the [PERIOD_END] (i.e. 23:59:59).

Examples:

2010-Q2, REPORTING_YEAR_START_DAY = --07-01 (July 1)

1. [REPORTING_YEAR_START_DATE] = 2010-07-01

b) [REPORTING_YEAR_BASE] = 2010-07-01

1. [PERIOD_DURATION] = P3M 3. (2-1) * P3M = P3M

2010-07-01 + P3M = 2010-10-01

[PERIOD_START] = 2010-10-01

4. 2 * P3M = P6M

2010-07-01 + P6M = 2010-13-01 = 2011-01-01

2011-01-01 + -P1D = 2010-12-31

[PERIOD_END] = 2010-12-31

The actual calendar range covered by 2010-Q2 (assuming the reporting year begins July 1) is 2010-10-01T00:00:00/2010-12-31T23:59:59

2011-W36, REPORTING_YEAR_START_DAY = --07-01 (July 1)

1. [REPORTING_YEAR_START_DATE] = 2010-07-01

a) 2011-07-01 = Friday

2011-07-01 + P3D = 2011-07-04

[REPORTING_YEAR_BASE] = 2011-07-04

1. [PERIOD_DURATION] = P7D 3. (36-1) * P7D = P245D

2011-07-04 + P245D = 2012-03-05

[PERIOD_START] = 2012-03-05

4. 36 * P7D = P252D

2011-07-04 + P252D =2012-03-12

2012-03-12 + -P1D = 2012-03-11

[PERIOD_END] = 2012-03-11

The actual calendar range covered by 2011-W36 (assuming the reporting year begins July 1) is 2012-03-05T00:00:00/2012-03-11T23:59:59

4.2.7 Distinct Range

In the case that the reporting period does not fit into one of the prescribe periods above, a distinct time range can be used. The value of these ranges is based on the ISO 8601 time interval format of start/duration. Start can be expressed as either an ISO 8601 date or a date-time, and duration is expressed as an ISO 8601 duration. However, the duration can only be positive.

4.2.8 Time Format

In version 2.0 of SDMX there is a recommendation to use the time format attribute to gives additional information on the way time is represented in the message. Following an appraisal of its usefulness this is no longer required. However, it is still possible, if required , to include the time format attribute in SDMX-ML.

Table 1: SDMX-ML Time Format Codes

4.2.9 Time Zones

In alignment with ISO 8601, SDMX allows the specification of a time zone on all time periods and on the reporting year start day. If a time zone is provided on a reporting year start day, then the same time zone (or none) should be reported for each reporting time period. If the reporting year start day and the reporting period time zone differ, the time zone of the reporting period will take precedence. Examples of each format with time zones are as follows (time zone indicated in bold):

• Time Range (start date): 2006-06-05-05:00/P5D
• Time Range (start date-time): 2006-06-05T00:00:00-05:00/P5D
• Gregorian Year: 2006-05:00
• Gregorian Month: 2006-06-05:00
• Gregorian Day: 2006-06-05-05:00
• Distinct Point: 2006-06-05T00:00:00-05:00
• Reporting Year: 2006-A1-05:00
• Reporting Semester: 2006-S2-05:00
• Reporting Trimester: 2006-T2-05:00
• Reporting Quarter: 2006-Q3-05:00
• Reporting Month: 2006-M06-05:00
• Reporting Week: 2006-W23-05:00
• Reporting Day: 2006-D156-05:00
• Reporting Year Start Day: 07-01-05:00

According to ISO 8601, a date without a time-zone is considered "local time". SDMX assumes that local time is that of the sender of the message. In this version of SDMX, an optional field is added to the sender definition in the header for specifying a time zone. This field has a default value of 'Z' (UTC). This determination of local time applies for all dates in a message.

4.2.10 Representing Time Spans Elsewhere

It has been possible since SDMX 2.0 for a Component to specify a representation of a time span. Depending on the format of the data message, this resulted in either an element with 2 XML attributes for holding the start time and the duration or two separate XML attributes based on the underlying Component identifier. For example, if REF_PERIOD were given a representation of time span, then in the Compact data format, it would be represented by two XML attributes; REF_PERIODStartTime (holding the start) and REF_PERIOD (holding the duration). If a new simple type is introduced in the SDMX schemas that can hold ISO 8601 time intervals, then this will no longer be necessary. What was represented as this:

 <Series REF_PERIODStartTime="2000-01-01T00:00:00" REF_PERIOD="P2M"/>

can now be represented with this:

 <Series REF_PERIOD="2000-01-01T00:00:00/P2M"/>

4.2.11 Notes on Formats

There is no ambiguity in these formats so that for any given value of time, the category of the period (and thus the intended time period range) is always clear. It should also be noted that by utilizing the ISO 8601 format, and a format loosely based on it for the report periods, the values of time can easily be sorted chronologically without additional parsing.

4.2.12 Effect on Time Ranges

All SDMX-ML data messages are capable of functioning in a manner similar to SDMXEDI if the Dimension at the observation level is time: the time period for the first observation can be stated and the rest of the observations can omit the time value as it can be derived from the start time and the frequency. Since the frequency can be determined based on the actual format of the time value for everything but distinct points in time and time ranges, this makes is even simpler to process as the interval between time ranges is known directly from the time value.

4.2.13 Time in Query Messages

When querying for time values, the value of a time parameter can be provided as any of the Observational Time Period formats and must be paired with an operator. This section will detail how systems processing query messages should interpret these parameters.

Fundamental to processing a time value parameter in a query message is understanding that all time periods should be handled as a distinct range of time. Since the time parameter in the query is paired with an operator, this also effectively represents a distinct range of time. Therefore, a system processing the query must simply match the data where the time period for requested parameter is encompassed by the time period resulting from value of the query parameter. The following table details how the operators should be interpreted for any time period provided as a parameter.

Reporting Time Periods as query parameters are handled like this: any data within the bounds of the reporting period for the year is matched, regardless of the actual start day of the reporting year. In addition, data reported against a normal calendar period is matched if it falls within the bounds of the time parameter based on a reporting year start day of January 1. When determining whether another reporting period falls within the bounds of a report period query parameter, one will have to take into account the actual time period to compare weeks and days to higher order report periods. This will be demonstrated in the examples to follow.

Examples:

Gregorian Period
Query Parameter: Greater than 2010
Literal Interpretation: Any data where the start period occurs after 2010-1231T23:59:59.
Example Matches:

• 2011 or later
• 2011-01 or later
• 2011-01-01 or later
• 2011-01-01/P[Any Duration] or any later start date
• 2011-[Any reporting period] (any reporting year start day)
• 2010-S2 (reporting year start day --07-01 or later)
• 2010-T3 (reporting year start day --07-01 or later)
• 2010-Q3 or later (reporting year start day --07-01 or later)
• 2010-M07 or later (reporting year start day --07-01 or later)
• 2010-W28 or later (reporting year start day --07-01 or later)
• 2010-D185 or later (reporting year start day --07-01 or later)

Reporting Period
Query Parameter: Greater than or equal to 2010-Q3
Literal Interpretation: Any data with a reporting period where the start period is on or after the start period of 2010-Q3 for the same reporting year start day, or and data where the start period is on or after 2010-07-01. Example Matches:

• 2011 or later
• 2010-07 or later
• 2010-07-01 or later
• 2010-07-01/P[Any Duration] or any later start date
• 2011-[Any reporting period] (any reporting year start day)
• 2010-S2 (any reporting year start day)
• 2010-T3 (any reporting year start day)
• 2010-Q3 or later (any reporting year start day)
• 2010-M07 or later (any reporting year start day)
• 2010-W27 or later (reporting year start day --01-01)⁵
• 2010-D182 or later (reporting year start day --01-01)
• 2010-W28 or later (reporting year start day --07-01)⁶
• 2010-D185 or later (reporting year start day --07-01)

4.3 Versioning

Versioning operates at the level of versionable and maintainable objects in the SDMX information model. Within the SDMX Structure and MetadataSet messages, there is a well-defined pattern for artefact versioning and referencing. The artefact identifiers are qualified by their version numbers – that is, an object with an Agency of "A", and ID of "X" and a version of "1.0.0" is a different object than one with an Agency of "A", an ID of "X", and a version of "1.1.0".

As of SDMX 3.0, the versioning rules are extended to allow for truly versioned artefacts through the implementation of the rules of the well-known practice called "Semantic Versioning" (http://semver.org), in addition to the legacy non-restrictive versioning scheme. In addition, the "isFinal" property is removed from MaintainableArtefact. According to the legacy versioning, any artefact defined without a version is equivalent to following the legacy versioning, thus having version ‘1.0’.

4.3.1 Non-versioned artefacts

Indeed, some use cases do not need or are incompatible with versioning for some or all their structural artefacts, such as the Agency, Data Providers, Metadata Providers and Data Consumer Schemes. These artefacts follow the legacy versioning, with a fixed version set to ‘1.0’.

Many existing organisation’s data management systems work with version-less structures and apply ad-hoc structural metadata governance processes. The new nonversioned artefacts will allow supporting those numerous situations, where organisations do not manage version numbers.

4.3.2 Semantically versioned artefacts

Since the purpose of SDMX versioning is to allow communicating the structural artefact changes to data exchange partners and connected systems, SDMX 3.0 offers Semantic Versioning (aka SemVer) with a clear and unambiguous syntax to all semantically versioned SDMX 3.0 structural artefacts. Semantic versioning will thus better respond to situations where the SDMX standard itself is the only structural contract between data providers and data consumers and where changes in structures can only be communicated through the version number increases.

The semantic version number consists of four parts: MAJOR, MINOR, PATCH and EXTENSION, the first three parts being separated by a dot (.), the last two parts being separated by a hyphen : MAJOR.MINOR.PATCH-EXTENSION. All versions are ordered.

The detailed rules for semantic versioning are listed in chapter 14 in the annex for “Semantic Versioning”. In short, they define:

Given a version number MAJOR.MINOR.PATCH (without EXTENSION), when making changes to that semantically versioned SDMX artefact, then one must increment the:

1. MAJOR version when backwards incompatible artefact changes are made,
2. MINOR version when artefact elements are added in a backwards compatible manner, or
3. PATCH version when backwards compatible artefact property changes are made.

When incrementing a version part, the right-hand side parts are 0-ed (reset to ‘0’).

Extensions can be added, changed or dropped.

Given an extended version number MAJOR.MINOR.PATCH-EXTENSION, when making changes to that versioned artefact, then one is not required to increment the version if those changes are within the allowed scope of the version increment from the previous version (if that existed); otherwise, the above version increment rules apply. EXTENSIONs can be used e.g., for drafting or a pre-release.

Semantically versioned SDMX artefacts will thus be safe to use. Specific version patterns allow them to become either immutable, i.e., the maintainer commits to never change their content, or changeable only within a well-defined scope. If any further change is required, a new version must be created first. Furthermore, the impact of the further change is communicated using a clear version increment. The built-in version extension facility allows for eased drafting of new SDMX artefact versions.

The production versions of identifiable artefacts are assumed stable, i.e., they do not have an EXTENSION. This is because once in production, an artefact cannot change in any way, or it must change the version. For cases where an artefact is not static, like during the drafting, the version must indicate this by including an EXTENSION. Draft artefacts should not be used outside of a specific system designed to accommodate them. For most purposes, all artefacts should become stable before being used in production.

4.3.3 Legacy-versioned artefacts

Organisations wishing to keep a maximum of backwards compatibility with existing implementations can continue using the previous 2-digit convention for version numbers (MAJOR.MINOR) as in the past, such as '2.3', but without the ‘isFinal’ property. The new SDMX 3.0 standard does not add any strict rules or guarantees about changes in those artefacts, since the legacy versioning rules were rather loose and non-binding, including the meaning of the ‘isFinal’ property, and their implementations were varying.

In order to make artefacts immutable or changes truly predictable, a move to the new semantic versioning syntax is required.

4.3.4 Dependency management and references

New flexible dependency specifications with wildcarding allow for easier data model maintenance and enhancements for semantically versioned SDMX artefacts. This allows implementing a smart referencing mechanism, whereby an artefact may reference:

• a fixed version of another artefact
• the latest available version of another artefact
• the latest backward compatible version of another artefact, or the latest backward and forward compatible version of another artefact.

References not representing a strict artefact dependency, such as the target artefacts defined in a MetadataProvisionAgreement allow for linking to all currently available versions of another artefact. Another illustrative case for such loose referencing is that of Constraints and flows. A Constraint may reference many Dataflows or Metadataflows, the addition of more references to flow objects does not version the Constraint. This is because the Constraints are not properties of the flows – they merely make references to them.

Semantically versioned artefacts must only reference other semantically versioned artefacts, which may include extended versions. Non-versioned and legacy-versioned artefacts can reference any other non-versioned or versioned (whether semantic or legacy) artefacts. The scope of wildcards in references adapts correspondingly.

The mechanism named "early binding" refers to a dependency on a stable versioned artefact – everything with a stable versioned identity is a known quantity and will not change. The "late binding" mechanism is based on a wildcarded reference, and it resolves that reference and determines the currently related artefact at runtime.

One area which is much impacted by this versioning scheme is the ability to reference external objects. With the many dependencies within the various structural objects in SDMX, it is useful to have a scheme for external referencing. This is done at the level of maintainable objects (DSDs, Codelists, Concept Schemes, etc.) In an SDMX Structure Message, whenever an "isExternalReference" attribute is set to true, then the application must resolve the address provided in the associated "uri" attribute and use the SDMX Structure Message stored at that location for the full definition of the object in question. Alternately, if a registry "urn" attribute has been provided, the registry can be used to supply the full details of the object.

The detailed rules for dependency management and references are listed in chapter 14 in the annex for “Semantic Versioning”.

In order to allow resolving the described new forms of dependencies, the SDMX 3.0 Rest API supports retrievals legacy-versioned, wildcarded and extended artefact versions:

• Artefact queries for a specific version (X.Y, X.Y.Z or X.Y.Z-EXT).
• Artefact queries for latest available semantic versions within the wildcard scope (X+.Y.Z, X.Y+.Z or X.Y.Z+).
• Queries for non-versioned artefacts.
• Artefact queries for all available semantic versions within the wildcard scope (*, X.* or X.Y.*), where only the first form is required for resolving wildcarded loose references.

The combination of wildcarded queries with a specific version extension is not permitted.

Full details can be found in the SDMX RESTful web services specification.

4.4 Structural Metadata Querying Best Practices

When querying for structural metadata, the ability to state how references should be resolved is quite powerful. However, this mechanism is not always necessary and can create an undue burden on the systems processing the queries if it is not used properly.

Any structural metadata object which contains a reference to an object can be queried based on that reference. For example, a categorisation references both a category and the object is it categorising. As this is the case, one can query for categorisations which categorise a particular object or which categorise against a particular category or category scheme. This mechanism should be used when the referenced object is known.

When the referenced object is not known, then the reference resolution mechanism could be used. For example, suppose one wanted to find all category schemes and the related categorisations for a given maintenance agency. In this case, one could query for the category scheme by the maintenance agency and specify that parent and sibling references should be resolved. This would result in the categorisations which reference the categories in the matched schemes to be returned, as well as the object which they categorise.

1. ^ Regular expressions, as specified in W3C XML Schema Definition Language (XSD) 1.1 Part 2: Datatypes.
2. ^ The seconds can be reported fractionally
3. ^ ISO 8601 defines alternative definitions for the first week, all of which produce equivalent results. Any of these definitions could be substituted so long as they are in
   relation to the reporting year start day.
4. ^ The rules for adding durations to a date time are described in the W3C XML Schema specification. See http://www.w3.org/TR/xmlschema-2/#adding-durations-todateTimes for further details.
5. ^ 2010-Q3 (with a reporting year start day of 01-01) starts on 2010-07-01. This is day 4 of week 26, therefore the first week matched is week 27.
6. ^ 2010-Q3 (with a reporting year start day of 07-01) starts on 2011-01-01. This is day 6 of week 27, therefore the first week matched is week 28.