Skip to content

Release candidate - #190

Draft
jwrosewell wants to merge 283 commits into
mainfrom
rc
Draft

jwrosewell wants to merge 283 commits into
mainfrom
rc

Conversation

@jwrosewell

@jwrosewell jwrosewell commented Nov 16, 2026 •

Copy link
Copy Markdown
Contributor

Releases what has collected on rc since the last release, #112, which merged on 13 November 2026 at 10:35 UTC. Maven Central refused that release's publish (run 43410313251, "Publishing blocked: your organization has reached its monthly release size limit ... Usage resets 2026-12-01"), so nothing of #112 was published and 4.5.130 is still the newest version of every package. The packages this release publishes therefore also carry every change #112 lists, and "Before you upgrade" below covers anyone moving from 4.5.130.

This release carries context7.json at the root, which tells Context7 how to list the repository. The file changes no package.

This release merges into main on Tuesday 17 November, and Publish runs on 1 December, when Maven Central's monthly limit resets. Only the patch number moves.

The C library. This release is built on the cxx submodule at 7992af70, the 4.5.106 release, read from the gitlink of the head. It is not built on this week's C library release. Moving Java forward is open, and What remains says so.

What this release brings into conformance with the specification

The Usage Sharing Element accepts no Evidence and reads the web request

usage-sharing-element.md lines 90 to 96 at dbaf099c says

The element accepts no Evidence, so it does not accept the session cookie either. When the track session option is on, the element MUST read the cookie named by the session cookie name option from the same two places as the values it shares, being the Evidence in the Flow Data and the web request supplied as server.http-request.

The element now accepts no Evidence and reads the values it shares from the Evidence in the Flow Data and from the web request. A test sends a header that no element reads through the web integration and finds it in the usage sharing message (#127), and further tests show that within one session a request differing only in its sequence is not shared again whilst a changed 51D_ value is, and that JA4, H2 and both billing keys added beside the web request reach the message (#136). Only the Cookie header is blocked by default, so a header named Cookies is shared (#132), and the filter compares Evidence keys in the root locale, so a Turkish or Azeri default locale no longer lets ID.EMAIL, HEADER.COOKIE or a COOKIE.<name> key past the rules that keep them out (#83). The record itself is written as the specification's Processing section sets out (#139), and closing stops sending after its wait, discards the rest and warns with the number of entries left (#147).

Usage sharing shares every query value by default, and no cookie but the 51D_ ones

usage-sharing-element.md lines 65 to 69 at 30b3741f says

It MUST also return false, whatever the configuration and even when all Evidence is shared, for any key whose prefix is cookie. and whose field does not start with 51D_.

and lines 76 to 81 say "Share all query string parameters is on by default, so every query string and HTTP form parameter is shared unless a user turns it off." Both are now the defaults (#160).

IP Intelligence reads its native results one thread at a time, types Latitude and Longitude as numbers, and processes in the cloud with the shared logic

ip-intelligence-on-premise.md lines 520 to 525 at dbaf099c says Aspect Data that reads from a ResultsIpi instance "MUST let only one thread at a time read from its results", data-model.md lines 48 and 49 give Latitude and Longitude the type float, and ip-intelligence-cloud.md lines 48 and 49 say "The majority of the processing SHOULD be handled by the shared Cloud Aspect Engine logic." IP Intelligence Element Data now holds one lock for every read of its native results, the property metadata types a native single as Float, and the IP Intelligence cloud engine leaves processing to the shared Cloud Aspect Engine logic (#177).

A property list splits at every comma, and Element Data is found whatever the case of its key

pipeline-configuration.md lines 50 to 61 at b3b2c0dd gives a list setting in a configuration file as one comma-separated string, and access-to-results.md line 11 at b3b2c0dd says "String comparisons MUST be case insensitive." An engine builder's setProperties(String) now splits the list at every comma and trims each name, where it split only at a comma with one white space character on each side (#166), and Flow Data finds Element Data by its key whatever the case, so flowData.get("Device") finds what is stored under device (#167).

A missing Property gives the specification's message and reason

properties.md lines 116 to 118 at dbaf099c says

When the user attempts to access a Property that is defined for an Engine, but does not exist in the result set for some reason, the system MUST throw an error/exception with a message explaining why the Property is not available.

and the table under it gives each reason's message. The missing Property messages are now that table's text, the reason is decided in the order the specification gives, a failed cloud request has its own reason, CloudRequestFailed, and the on-premise Device Detection and IP Intelligence data read each Property by its position rather than asking the native results by name (#175). The same change turns usage sharing off in the IP Intelligence console examples with the comment the specification asks for, adds a programmatic data update example, ManualDataUpdate, declares what FlowData throws, and deprecates the test utility classes FileUtils and KeyUtils in the published device-detection.shared jar.

The JSON Builder writes errors as a list of messages

json-builder.md lines 106 to 107 at dbaf099c says

Add any errors from the Flow Data errors collection to the result. This is structured as a string array of the error messages under the top-level key errors.

errors is now a list of messages in the order they were recorded, where it was an object keyed by element (#128). A Property assigned no value is written as null beside its nullreason, at the top level and in each item of a list of Element Data, and the JavaScript Builder's public constructors refuse a protocol that is neither http nor https (#145).

The JavaScript Builder minifies its script, with a library, and its cost is tested

javascript-builder.md lines 105 to 114 at dbaf099c says

Finally, the resulting JavaScript will be minified (or not, based on the configuration provided). Minification SHOULD be performed using an existing third-party library.

Performance tests MUST be in place to verify the performance impact of this feature.

and line 350 gives SetMinify the default True. The script was never minified, whatever SetMinify said. It is now minified with net.logicsquad:minifier once for each form of the template rather than on each request, with each request's values put into the minified form afterwards so that the minifier never touches them, a performance test holds a minified render to within twice the cost of an unminified one, and a test reads the JSON endpoint's answer and the JSON in the script and holds them to the same form (#178).

A cloud engine works with empty Property metadata

cloud-aspect-engine.md lines 57 to 61 at dbaf099c says

Where the Property metadata is empty, for example before discovery has completed, a Cloud Aspect Engine MUST still work, reading the response JSON and inferring the type of each value from the value itself, as it already does, and anything that reads the metadata to decide what the Engine can produce MUST tolerate finding nothing there.

The Device Detection cloud engine gave none of the response's values when its metadata was empty, and reading one threw PropertyMissingException. It now reads them, with a number given the type its typed accessor declares, and the Cloud Request Engine logs each entry of a response's warnings (#172).

The client-side endpoints answer a revalidation without running the Pipeline

web-integration.md lines 214 to 222 at dbaf099c says

The Pipeline Evidence key filter is the union of the accepted Evidence of the Pipeline's Flow Elements, and a Flow Element accepts only the Evidence that can change its results, so the ETag and the Vary header cover only those keys. The Usage Sharing Element accepts none. If-None-Match changes no result, so no Flow Element accepts it, and a revalidation's ETag is the same as the one it sends.

The JavaScript and JSON endpoints send a weak ETag of the Evidence the request supplied and answer a matching If-None-Match before the Pipeline runs, in the servlet filter and in the Spring interceptor, with a 304 to a GET or HEAD and a 412 to any other method but CONNECT, OPTIONS and TRACE. The 304 carries the Vary, Cache-Control, ETag and CORS headers of a good response, a response built whilst processing recorded errors sends Cache-Control: private, no-cache and no ETag, and the maximum age can be set, with 1800 seconds still the default (#67). The web integration also supplies server.request-query, the query string as sent and empty when there is none, which usage sharing never shares (#144).

A data update failure is logged as its row of the Errors table

data-updates.md line 481 at dbaf099c says of the Errors table's messages

These messages MUST also be consistent across languages to make life easier for 51Degrees support.

Every HTTP failure of the data file download, one cut short included, is logged as its row of that table, and a failed check is logged at Error with the reason in the attached error (#129). Four more rows now match the table: a response with no Content-MD5 header is a warning and the data is applied (data-updates.md line 505), a failed integrity check gives AUTO_UPDATE_ERR_MD5_VALIDATION_FAILED with both hashes in its message, a data file with no temporary file path gives the new AUTO_UPDATE_TEMP_PATH_NOT_SET, and a failure writing data passed in from memory is a warning with the engine still refreshed (#186).

A cloud Property the response leaves out is reported, not guessed

The Device Detection and IP Intelligence cloud engines now throw PropertyMissingException for a Property the response leaves out, as Location already did, and a Property the response gives only as a nullreason keeps that reason (#143). The cloud Pipeline builders take the properties wanted, which a caller authenticating with a licence key alone needs, and a live test authenticates on a licence key alone (#131).

51Did factor words are read whatever their case

Factor words are compared ignoring case, so a redeem answer's Verified or NotRecorded reads as that outcome where it read as Mismatch, and a payload of one to four bytes naming version 1, 2 or 3 is refused as UNSUPPORTED_PAYLOAD_VERSION naming that version, where it was PAYLOAD_TOO_SHORT (#141).

Before you upgrade

  1. Each package now has its own version. Packages are released only when a file under them changed since their last release, or a package they depend on is released, and each is tagged <package>/<version> (#63). A project that sets one version for every 51Degrees artifact needs each artifact's own version from now on. Each published module requires the versions it was built with, so Maven resolves a consistent set from any one of them. The Build section gives the plan the Publish dry run printed, package by package.
  2. The JSON Builder writes errors as a list of messages, in the order the errors were recorded, where it wrote an object keyed by element. An application that reads the JSON Builder's output and treats errors as an object must read it as a list, which is what the specification asks for, the .NET builder writes and the client-side script reads (#128).
  3. The JSON Builder writes a Property assigned no value as null, beside its nullreason, at the top level and in each item of a list of Element Data, where it wrote only the reason. Code that took a missing key to mean no value now finds the key with a null value (#145).
  4. A JavaScriptBuilderElement made with a public constructor refuses a protocol that is not http or https, white space included, with PipelineConfigurationException when it is made. Before, the element was made and gave a script with no callback URL on each request. The builder already refused such a protocol (#145).
  5. An application that builds its own Evidence has to give Usage Sharing the web request. This applies to an application that adds a value to Evidence only when Pipeline.getEvidenceKeyFilter() includes its key. An application that uses the web integration has nothing to change. Up to 4.5.130 Usage Sharing accepted the headers it shares, so that check kept them. Usage Sharing now accepts none, so the same check drops every header that no other Flow Element reads, and those headers stop being shared. Add the request with flowData.addEvidence(ShareUsageEvidence.HTTP_REQUEST_KEY, request). ShareUsageEvidence is in fiftyone.pipeline.engines.fiftyone.data, and the key is server.http-request. The value is a java.util.Map that holds each value of the request under the key it would have as Evidence, such as header.user-agent, cookie.51D_Profile, query.q and server.client-ip. Usage Sharing shares every Evidence value in the Flow Data, then each value of that map whose key the Evidence does not hold. Its rules still apply, so the Cookie header, a blocked header and a cookie whose name does not start 51D_ are left out. The map itself is never shared and is not sent to the cloud service. In a servlet application, WebRequestEvidenceServiceCore.addEvidenceFromRequest in pipeline.web adds the request for you (#68).
  6. A request header named Cookies is now shared by Usage Sharing by default, as every other header but Cookie is. To keep it out, add it to the blocked headers with setBlockedHttpHeader("Cookies") on the usage sharing builder (#132).
  7. The client-side endpoints send a weak ETag of the Evidence the request supplied, and answer a matching If-None-Match before the Pipeline runs. An ETag kept from an earlier version does not match, so each browser gets one full response after the upgrade. An application that wires the web integration itself, rather than through PipelineFilter or FiftyOneInterceptor, calls the new serveNotModified before processing to answer revalidations the same way (#67).
  8. A failed check for a data update is logged at Error as "Error during check for data update", with the reason in the attached error, where it was a warning. An application that alerts on Error entries sees these (#129).
  9. A cloud Property the response leaves out throws PropertyMissingException from the Device Detection and IP Intelligence cloud engines, where it gave a value of Unknown or null. Location behaved this way already (#143).
  10. Usage sharing's close gives up waiting after its limit and discards what is left in the queue, with a warning giving the number of entries left, rather than holding the process that closes it (#147).
  11. The usage sharing record changes form. Its root no longer carries version="1.1", it is sent as text/xml where it was text/xml; charset=utf-8, FlowElement holds each element's fully qualified class name, and SessionId, Version, LanguageVersion, ClientIP, ServerIP and Platform are written empty where they were left out, each as the specification's Processing section sets out (#139). Anything that reads the record itself, rather than the usage sharing service, sees the new form.
  12. Missing Property reasons and messages change (#175). The messages are the specification's text. A Property missing because the cloud request failed gives the new reason CloudRequestFailed. A Cloud Engine no longer gives DataFileUpgradeRequired, and one whose metadata has not loaded gives Unknown or CloudRequestFailed where it gave ProductNotAccessibleWithResourceKey. Code that matched on a message's text or on those reasons needs to follow.
  13. A Property named when the on-premise engine was built, and missing from the data file, now reads as a value with hasValue() false and the native message, where reading it threw PropertyMissingException (#175). This follows from reading each Property by its position. It is read from the code and has no test of its own, and whether such a read should throw with the upgrade message instead is an open question on the specification.
  14. FlowDataDefault.get(String) before processing throws IllegalStateException, a subclass of the RuntimeException it threw, so a caller that catches RuntimeException sees no change (#175).
  15. FileUtils and KeyUtils in device-detection.shared are deprecated. They are test utility classes and will leave the published jar in a later release. The examples use their own DataFileHelper and KeyHelper (#175, #187).
  16. The client-side script is now minified by default. Up to 4.5.130 setMinify had no effect and the script was served as the template produced it. Anything that compares the script byte for byte, or reads it expecting the template's layout, sees minified output. Set setMinify(false) on the JavaScript Builder to keep the old output (#178).
  17. pipeline.javascriptbuilder has a new dependency, net.logicsquad:minifier 1.0, under the 2-Clause BSD licence, with its log4j binding for SLF4J left out so that it does not take over an application's logging. A project that pins its own transitive versions needs it on the path (#178).
  18. Usage sharing shares every query string and HTTP form parameter by default. To share only the 51D_ parameters and those named, as before, set setShareAllQueryStringParameters(false) on the builder, or ShareAllQueryStringParameters to false in the options file (#160).
  19. With share all Evidence on, a cookie whose name does not start with 51D_ is no longer shared, the session cookie included. There is no setting to share it (#160).
  20. Latitude and Longitude are Float values where they were String values wherever the type is taken from the metadata: getType(), get("latitude"), get("longitude") and asKeyMap() on the on-premise data, and so the JSON the JSON Builder writes for them, which now carries numbers where it carried text (#177).
  21. The IP Intelligence cloud engine behaves as the Device Detection one does when the cloud request fails. With no response it logs the shared warning and gives no values, where it failed reading the response, and with no Cloud Request Engine before it in the Pipeline it throws PipelineConfigurationException. A subclass that overrode processEngine overrides processCloudEngine instead (#177).
  22. A data update response with no Content-MD5 header is now applied, with a warning, where it stopped the update with AUTO_UPDATE_ERR_MD5_VALIDATION_FAILED. A data file with no temporary file path reports AUTO_UPDATE_TEMP_PATH_NOT_SET, a new member at the end of AutoUpdateStatus, where it reported AUTO_UPDATE_REFRESH_FAILED, and a failure writing data passed in from memory is logged at Warning, not Error (#186).
  23. setProperties("IsMobile,BrowserName") now asks for two properties, where it asked for one named IsMobile,BrowserName (#166).
  24. Element Data keys are compared ignoring case. flowData.get("Device") finds the Element Data stored under device, where it threw NoSuchElementException, and adding Element Data under a key that differs only in case finds what is already there (#167).

New features

  • A programmatic data update example, ManualDataUpdate, which asks the data update service for an update now (#175).
  • Package descriptions say what each package does in the words people search for, and the parent poms' source links point at this repository (#27).
  • The web integration supplies server.request-query (#144).
  • The builder defaults that were literals are named constants, with no value changed, and the lazy loading Javadoc gives the real default of 1 second (#130).
  • Every Device Detection and IP Intelligence example ends with a Find out more section, tagged for the example, in the console and on the web pages, an unknown performance profile ends the same way, the Device Detection web cloud options file uses https, and the IP Intelligence web examples no longer fall back to the Enterprise file (#133). The Pipeline developer examples and the Location console examples do the same (#134), and the IP Intelligence API example says at start-up what the Enterprise data file adds, with the contact link (#159).

Performance and test coverage

  • The share usage overhead test measures a median over warmed up rounds, with limits set from every runner's figures (#91).
  • Chrome starts once before the Selenium contract tests (#124), and so does the first Chrome driver session, timed, so the first test no longer carries the slow first start on the Intel macOS runners (#138).
  • A test checks the positions of the elements the web integration adds (#155), tests keep FodId's byte layout, payload version and field-setting constructor out of the public surface (#157), and the configured interval data update test no longer races when its random delay comes out as zero (#161).
  • The usage sharing builder tests, which JUnit 5 passed over because they were written for JUnit 4, now run (#142).
  • A workflow builds and runs the getting started examples from a fresh clone against the packages on Maven Central, on each runner's own JDK, downloading each free data file once a run (#137).

Housekeeping

  • Each pull request builds on the configurations its changes need, and the CI script tests run on pull requests (#43).
  • Configuration page links point at https://51degrees.com/configure (#115), AGENTS.md points at common-ci's one list of environment variable names (#118), AGENTS.md records that a test failing on timing is fixed rather than only re-run (#122), and AGENTS.md and the example READMEs describe the examples as they run today (#162).

Fixed

Issues this closes

Build

The four release checks run on the final head between 02:00 and 05:00 UTC on Tuesday 17 November, being the release build with macOS, the Publish dry run, the Nightly Pipeline dry run and the timing against main. This section gives each run and its result then, and nothing in this body is reported as checked until it does.

Independent review

Every change in this release was reviewed as its own draft into rc at the time it merged. This release body has not yet been independently reviewed at its head. A review of rc 20f9089 against main f212dfc on 16 November left the release not ready on three findings, which were that no release draft existed, that no release check had run on the head, and that the examples did not build from a fresh clone. This draft answers the first, the Build section answers the second once the checks run, and #187 fixed the third.

What remains

  1. The C library. This release is built on cxx 7992af70, the 4.5.106 release, so the weighted Latitude and Longitude are not rounded to 3 decimal places, which this week's C library release does. Whether Java moves to it is open.
  2. The package build on the runner's newest JDK. The scheduled Nightly Pipeline on main fails in its Build job on ubuntu-26.04 with "Unsupported class file major version 69", which is a Java 25 class file read by PMD 3.26.0. The fix, #140, is not in this head, so until it is the dashboard badge stays red and the Publish and nightly dry runs stop in Build.
  3. Examples. A cloud example run with no resource key ends with an exception (#163), Comparer needs the Enterprise data file (#158), and a missing Property is answered with an HTML link printed to the console (#193). None of those is in this head.
  4. Merging this release into main starts a real publish at the next 03:00 UTC Nightly Pipeline, because its Publish job runs with dryrun false on the schedule. Until #140 is in, that job stops in Build and uploads nothing. Once it is in, each night's run tries the upload, which Maven Central refuses until its monthly limit resets on 1 December, so the nightly fails until then, and the first scheduled run after the reset publishes whatever main holds.
  5. The configuration page link in two missing Property messages. pipeline.engines' Constants still gives https://configure.51degrees.com in the messages for a Property not in the resource key, where #115 moved every other link to https://51degrees.com/configure, because #175 copied the specification's table, which still gives the old address.
  6. The issues left open and why each cannot be done for this release are listed with the release.

…ends logged, the repeat time, the pause messages and processing during a pause
… license key alone against the live service (#131)

Closes #12.
… blocked (#132)

The usage sharing filter blocks two headers by default where the specification blocks one. `EvidenceKeyFilterShareUsage` adds both `cookies` and `cookie` to its blocked headers whatever it is given ([lines 165 to 168 at rc 1c99b72](https://github.com/51Degrees/51degrees-java/blob/1c99b72f975209ab30082985f6cceeaf71d9aca1/pipeline/pipeline.engines.fiftyone/src/main/java/fiftyone/pipeline/engines/fiftyone/data/EvidenceKeyFilterShareUsage.java#L165-L168)), so a request header named `Cookies` is never shared unless all Evidence is shared. `Cookies` is not the Cookie header, so it is an ordinary header that the specification shares by default.
CsvNumbersTest expects the distance CompareOnPrem writes into its CSV
and the average MetricsOnPrem writes into its CSV with a full stop under
a German default locale. RandomIpTest expects the client IP addresses
ShareUsageExample makes up to use the digits 0 to 9 under a Saudi
Arabian default locale.
CompareOnPrem and MetricsOnPrem wrote a distance and an average into
their CSV rows with String.format and the default locale, so a locale
with a decimal comma split one column into two. ShareUsageExample made
up client IP addresses the same way, so a locale whose digits are not 0
to 9 gave Evidence no engine can read. Each now formats in Locale.ROOT,
and the two CSV numbers are package-private methods so a test can reach
them.
… 9110 ask (#67)

Closes #11. This brings the client-side JavaScript and JSON endpoints of the web integration, served by both the servlet filter and the Spring interceptor, in line with the [Cache header detail](https://github.com/51Degrees/specifications/blob/6570727d2af6967390d3c4fbec77b9cf0540d3f3/pipeline-specification/features/web-integration.md#cache-header-detail) section of the web integration specification, quoted below at [6570727d](51Degrees/specifications@6570727), the head of the specifications `main` branch. That section was written by [51Degrees/specifications#53](51Degrees/specifications#53), [#81](51Degrees/specifications#81), [#115](51Degrees/specifications#115), [#121](51Degrees/specifications#121) and [#136](51Degrees/specifications#136), all merged.
…s the specification's Errors table rows (#129)

Closes #125.
…t more section (#133)

Three gaps in the Device Detection and IP Intelligence examples, found in the audit of the examples on 13 November.
… out more section (#134)

No Pipeline developer example and no Location example ended with a "Find out more" section, found in the audit of the examples on 13 November. The 51Degrees rule for examples is that every example ends with a clearly separated "Find out more" section that gives the person running it links to follow up, being the relevant repositories as plain addresses and 51degrees.com pages tagged with the campaign convention that common-ci's [`scripts/utm-lint.ps1`](https://github.com/51Degrees/common-ci/blob/1c5c0e0b09a1f77722e20ba2d10cbd1031caca9e/scripts/utm-lint.ps1) checks.
The web integration supplies the method and the path of the request line as Evidence, and not the query string. [web-integration.md lines 95 to 114 at 706de1f0](https://github.com/51Degrees/specifications/blob/706de1f0cb3edd560eb7fd7ef3bb19b93a4883dc/pipeline-specification/features/web-integration.md?plain=1#L95-L114) list "Request query string, whole and undecoded. Key is `server.request-query`" beside the method and the path, and say:
ExamplesWithoutSettingsTest.comparer runs Comparer's main with no
arguments, as from a fresh clone, and expects both solutions to report
and the note on what the Enterprise data file adds, with the contact
link. ComparerTest is removed, because it skipped whenever the
Enterprise data file was missing, which is every CI run, and its run is
now this one.
…file adds

Comparer stopped with 'Enterprise data file not found' on a fresh clone,
as it needed Enterprise-HashV41.hash, which no clone has. The 51Degrees
solution now uses the free Lite data file, downloaded if it is missing,
unless the first argument names another data file, as the other
examples do. On the Lite data file it asks for no property list, the
properties the file lacks read as needing the Enterprise data file, and
after the report the example says what the Enterprise data file adds,
with the tagged contact link. The class Javadoc no longer links the
pricing page.
…col in the JavaScript Builder's constructor (#145)

Two places where the JSON Builder and the JavaScript Builder differ from the specification, found by a conformance check of this repository against [specifications at 706de1f0](https://github.com/51Degrees/specifications/tree/706de1f0cb3edd560eb7fd7ef3bb19b93a4883dc).
Each cloud example runs in a new Java process with every variable whose
name holds KEY left out of its environment, as it runs from a fresh
clone, and must end with exit code 0, name the variable to set, link to
the configuration page with the campaign tags, and end with its Find out
more section. ExampleProcess, in each product's examples shared module,
starts the process, because a running Java process cannot remove a
variable from its own environment.
A cloud example run with no resource key threw an exception, failed in
its builder or, for the 51Did demo server, exited with status 1, and the
Location examples asked for a key to be pasted into the source. The
specification now says a cloud example run with no resource key ends
without an error, names the environment variable to set, gives the
tagged link to where a key is created, and exits successfully, so that
its test runs it with no settings.

Each Device Detection, IP Intelligence and Location cloud example, and
the 51Did creator context demo server, now looks for its key as its
argument and then in _51DEGREES_RESOURCE_KEY (the Device Detection
examples then in TestResourceKey, the TAC and native model examples in
SuperResourceKey, as before). With none it says which variable to set,
links to the configuration page tagged for the example, adds the
contact page for the TAC and native model examples, whose properties
need a paid licence, and ends with its Find out more section and exit
code 0. The web examples end the same way rather than start a server
that cannot answer.
…hort payload (#141)

Two places where the 51Did package differs from the specification, found by a conformance check of this package against [specifications at 706de1f0](https://github.com/51Degrees/specifications/tree/706de1f0cb3edd560eb7fd7ef3bb19b93a4883dc).
…#137)

Two gaps against the examples rule, that every example builds and runs both from a fresh clone and against the published packages, found in the audit of the examples on 13 November.
Tests only. The web integration adds the Sequence Element, the JSON Builder Element, the JavaScript Builder Element and the Set Headers Element where the application's Pipeline lacks them, in the positions the specification gives, but the only test checked that each one is there, not where. This adds a test of the positions.
…nstructor out of reach (#157)

Tests only. Three rules of the 51Did specification limit what `FodId` offers a caller, and the Java package meets all three, but no test would fail if a later change broke one of them. This adds those tests. It is the follow-up that [#141](#141) names under What remains.
…ence API example (#159)

The IP Intelligence API example answers in JSON, so on the free Lite data file a caller saw the three registration properties and was never told what the Enterprise data file adds, where to ask for it, or where to find out more.
…its verify (#161)

Tests only. `DataUpdateServiceTests.DataUpdateService_Register_AutoUpdateConfiguredInterval` fails about once in 600 runs, and it failed `Ubuntu_Java_17` in full build [43667464820](https://github.com/51Degrees/51degrees-java/actions/runs/43667464820) of [#156](#156), which changes nothing in `pipeline.engines`. This fixes the cause rather than leaving it to a re-run, which is the rule for a test that fails on timing. [#122](#122), which is open and not yet merged, adds that rule to AGENTS.md.
…ows will not write a file a reader has mapped
jwrosewell and others added 18 commits November 19, 2026 07:28
Adds context7.json, so Context7 lists the repository with one title
pattern and a description of its four features, states the EUPL 1.2 and
its notices in the rules, and leaves out the folders and files that are
not documentation for someone using the packages.
…iles

Take the IP Intelligence example map tiles from OpenStreetMap
The packages, examples, tests and CI read only the current _51DEGREES_
names. The fallbacks to the old names, their once-only messages and the
tests that existed only for the old names are removed.

The deprecated public constants that hold an old name stay, so code that
uses them still compiles, but the variables they name are no longer read.
The name check allows an old name only in those constants and their tests.
Stop reading the deprecated environment variable names
Log and report a failed data update check by polling or the file watcher, and go on checking
…n-text

Answer a missing Property in the examples with the contact link in plain text
Compare the JavaScript Builder's Promise value as text, so a browser with promises gets the promise form
Examples that build from a fresh clone, checked in CI
…in-timeout

Limit the cloud host lookup to the timeout and log why processing failed
Add the Metrics component and the profiles for a property value, and read Device Detection results by position with no lock
Make the cloud mixed web example pass the shared browser tests. Reviewed READY at 5d34c0f, with all 8 checks on that head passing.
The Device Detection and IP Intelligence on-premise engines now hold their own monitor for the whole of either refreshData overload. A second refresh of the same engine waits for the first to finish instead of running alongside it in the native code, and processing takes no lock. Green run 44652969918.
The description in context7.json now says how to use the Java library, naming the Maven artifacts, the Pipeline builder calls, and the Flow Data calls. It also points to the IP Intelligence builder, the web integration classes, and the 51Did client. Nothing else in the file changes. Green run 44570340481.
The Sequence Element's class comment now says it adds Evidence, so it must not run in a parallel group, and that the Pipeline does not check this. New tests confirm a Pipeline holding it in a parallel group still builds, from code and from configuration. Green run 44570340481.
The usage sharing element now makes a version 4 UUID when it is built.
It writes it as ProcessId, first in every record. Version, Product,
FlowElement, Package, Language, LanguageVersion, ServerIP and Platform
go in the first record, then again once an hour has passed. Other
records leave them out. There is one Package per 51Degrees package in
use, named by Maven group and artifact, with its version as the text.
Once process() returns or throws, adding Evidence, Element Data, or errors to a Flow Data throws IllegalStateException, including through getEvidence() and through any map from elementDataAsMap(). Reads and lazy properties still work, no lock is taken, and the FlowData comments now say a Flow Data is used by one thread at a time. Green run 44675448209.
51DTechResponse and others added 11 commits November 20, 2026 18:05
Listen on port 5151 by default in the web examples. Reviewed READY at dd92c7f, with all 8 checks on that head passing.
…use a refresh after close (#215)

Close an on-premise engine only after a running refresh ends, and refuse a refresh after close. Reviewed READY at cb4626d, with all 4 checks on that head passing.
Specification conformance, 51degrees-java, section 13. Reviewed at da25456 with no high finding open, and all 9 checks on that head passing.
#205)

Usage sharing now sends a query string or form parameter only when its name starts with 51D_ or a Flow Element in the Pipeline accepts it, even when all Evidence is shared. The included parameters list and the share all parameters option still compile but have no effect, and the repeat Evidence check hashes the same parameters the record sends. The Pipeline's Evidence key filter is now published through a volatile field. Green run 44570340481.
Add the Derived Property Element and its example. Reviewed at ae74210 with no high finding open, and all 6 checks on that head passing.
…, Evidence keys and the JavaScript Builder (#170)

Nine fixes the specification asks for, in usage sharing, data updates, Evidence keys and the JavaScript Builder. Reviewed at a19de29 with no high finding open, and all 8 checks on that head passing.
…it (#218)

Read each test key only under its _51DEGREES_ name, and skip without it. Reviewed at 7470fd3 with no high finding open, and all 11 checks on that head passing.
…make lazy loading have no effect (#202)

Fail a write to Element Data once its Flow Element has finished, and make lazy loading have no effect. Reviewed at 0af9bf2 with no high finding open, and all 5 checks on that head passing.
The watcher tests wait up to 30 seconds, three of macOS's 10 second polls, and stop as soon as the change arrives. The live cloud tests' engines get a 15 second timeout. Tests only.

Head c35cc6e.
…rns without waiting (#208)

Make the cloud response a lazy task, so the Cloud Request Engine returns without waiting. Reviewed at 47c3fba with no high finding open, and all 5 checks on that head passing.
Reuse Element Data through an object pool. Reviewed at 2cb7641 with no high finding open, and all 4 checks on that head passing.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment