Apply `XmlNameProcessor` consistently to all written element/attribute names by Sahana2524 · Pull Request #918 · FasterXML/jackson-dataformat-xml

Sahana2524 · 2026-09-27T15:27:13Z

Collection wrapper element name skips the XmlNameProcessor

startWrappedValue() writes the wrapper element straight to the Stax writer, so with a non-passthrough processor the wrapper name comes out verbatim while the wrapped item names (which go through writeName()) are encoded. Since the reader decodes every element name, a default wrapped List serialized with newAlwaysOnBase64Processor() then throws StreamReadException on read (the raw wrapper name is not valid base64), and a wrapper name holding an XML-invalid character such as a space emits non-well-formed XML. Passthrough output is unchanged.

Scope grew during review: names encoded at a single point

The same class of bug existed for other names that reached the Stax writer without going through the processor, so the fix was generalized:

Single encoding point. ToXmlGenerator._encodeName() is now the one place names are encoded. setNextName() / setNextNameIfMissing() encode their argument, as do writeName() and the wrapper name in startWrappedValue(). _nextName always holds the name as written.
Root element names (class name, @JsonRootName, ObjectWriter.withRootName()) are now encoded; previously e.g. a class Point broke the AlwaysOn base64 round trip. tofix/RootNameEncodingTest moved to misc/. The default namespace set in _initWithRootName() follows the encoded name.
Root attributes added via XmlGeneratorInitializer.addRootAttribute() are now encoded, so they can be read back like other attributes.
Standard-namespace names are never encoded or decoded. Names in the xsi, xml and xmlns namespaces are a fixed vocabulary: _encodeName() leaves them as-is, and XmlTokenStream._decodeName() no longer decodes xsi/xml names (StAX never reports xmlns declarations as attributes). Literal attributes such as xsi:noNamespaceSchemaLocation or xml:lang, and namespace declarations added as root attributes, stay usable by other XML tools and no longer fail with a name-decode error under base64 processors. writeName() also no longer carries these namespaces over from a preceding property to content names (Map / @JsonAnyGetter keys), which would have skipped their encoding.
Each name encoded once per write. The generator remembers the unencoded name next to _nextName, so setNextName() followed by writeName() of the same name does not encode twice, and namespaces are not processed twice; collection items reuse the encoded wrapper name when both are the same. This shortcut only applies while _nextName is still the exact name encoded from it, so direct assignments (subclasses, internal restores) cannot leave it stale. Bean serializers pass the unencoded property name and let the following writeName() encode it, so properties that end up not written (nulls, filtered out) are never encoded. The stream write context tracks unencoded names.

API / behavior changes (3.3.0, unreleased)

ToXmlGenerator.encodeContentName() (public, added in Apply XmlNameProcessor to unwrapped ObjectNode root element name #887, never released) is replaced by protected _encodeName(). Callers now always pass unencoded names, which removes the risk of double encoding.
With a non-passthrough XmlNameProcessor, names passed to setNextName() / setNextNameIfMissing() by custom serializers, and root attribute names, are now encoded. Before, they were written verbatim and could not be read back.
Names in the xsi, xml and xmlns namespaces (including bean properties explicitly declared in them) are no longer encoded on write or decoded on read. Content names following such a property no longer inherit its namespace. Compatibility note: with the AlwaysOn base64 processor, 3.2 wrote bean properties explicitly declared in the xsi or xml namespace with encoded names (e.g. xml:bGFuZw); 3.3 reads such names literally, so those properties will not match when reading XML written by 3.2. (The plain Base64 processor leaves valid names unchanged, so it is not affected for them.)

Tests (in XmlNameEscapeTest and RootNameEncodingTest)

AlwaysOn base64 round trips: wrapped collection (compact and pretty-printed), a non-base64-safe root name, and names set directly on the generator (setNextName(), startWrappedValue(), writeRepeatedPropertyName()).
Root attribute round trip, with xsi:, xml:lang and xmlns: attributes left literal; reading works with and without SKIP_UNKNOWN_XSI_ATTRIBUTES.
A content name after an xsi-namespaced property is encoded (Replacement processor).
Replacement processor output read back with the configured mapper, so the replaced wrapper must still be read as a wrapper.
A counting processor checks that each name is encoded exactly once per write (including a namespaced root and wrapped collections), and that unwritten properties are not encoded.
setNextName(null) still clears the pending name.

🤖 Generated with Claude Code

github-actions · 2026-09-27T15:28:27Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.70%	📈 +0.010%
Branches	70.97%	📈 +0.000%

github-actions · 2026-09-27T15:29:39Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.70%	📈 +0.010%
Branches	70.97%	📈 +0.000%

github-actions · 2026-09-28T22:16:51Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.70%	📈 +2.750%
Branches	70.97%	📈 +1.470%

github-actions · 2026-10-06T03:20:39Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.70%	📈 +0.010%
Branches	70.97%	📈 +0.000%

github-actions · 2026-10-06T03:32:35Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.70%	📈 +0.010%
Branches	70.97%	📈 +0.000%

github-actions · 2026-10-06T03:34:20Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.70%	📈 +0.010%
Branches	70.97%	📈 +0.000%

Sahana2524 · 2026-10-23T09:00:46Z

any update?

github-actions · 2026-10-26T00:52:48Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.70%	📈 +0.010%
Branches	70.97%	📈 +0.000%

Root names from class name, @JsonRootName or withRootName() were written without going through the XmlNameProcessor, so AlwaysOn base64 round trips failed for most POJOs. Encode in _initWithRootName() (covering all root paths, including unwrapped ObjectNode) and move RootNameEncodingTest out of tofix. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions · 2026-10-26T01:00:03Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.71%	📈 +0.020%
Branches	70.97%	📈 +0.000%

…asterXML#918) Names passed via setNextName()/setNextNameIfMissing(), writeName() and the startWrappedValue() wrapper now all go through one helper, _encodeName(), so _nextName always holds the encoded name. Fixes item names set via startWrappedValue() (used with writeRepeatedPropertyName()) being written unencoded, and avoids double encoding by replacing the (unreleased) public encodeContentName() with protected _encodeName(). Reuses existing QName when processor leaves name unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions · 2026-10-26T01:05:22Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.78%	📈 +0.090%
Branches	71.09%	📈 +0.120%

…h configured mapper - Add indented variant of AlwaysOn base64 wrapped collection round trip, covering wrapper start element written via XmlPrettyPrinter - Read Replacement processor output back with the configured mapper into a DTO declaring replaced names, verifying wrapper is recognized on read - Import JacksonXmlElementWrapper; reference FasterXML#918 in test comments Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions · 2026-10-26T01:07:24Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.78%	📈 +0.090%
Branches	71.09%	📈 +0.120%

…asterXML#918) Keep the unencoded name next to _nextName (_nextNameRaw), so that: - writeName() does not re-encode the name a bean serializer already set via setNextName() (each name now encoded once per write, not twice) - namespace carried over by writeName() is the unencoded one - writeRepeatedPropertyName() registers unencoded name in stream context, same as writeName() Also: - Encode root attribute names (except XSI ones meant for other tools) - Set default namespace from encoded root name in _initWithRootName() - Preserve prefix when re-creating encoded QName; simplify _encodeName() - Correct _nextName javadoc (fixed names are not encoded) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions · 2026-10-26T01:15:46Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.87%	📈 +0.180%
Branches	71.16%	📈 +0.190%

… names (FasterXML#918) - Only trust _nextNameRaw while _nextName is still the QName encoded from it (_nextNameEncoded), so direct assignments to _nextName (subclasses, internal restores) cannot leave a stale raw name; drops manual resets - XSI namespace names are standard names: ToXmlGenerator._encodeName() leaves them as-is and XmlTokenStream._decodeName() no longer decodes them, so literal xsi:* attributes (e.g. xsi:noNamespaceSchemaLocation) no longer fail to decode with base64 name processors Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions · 2026-10-26T01:27:09Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.89%	📈 +0.200%
Branches	71.22%	📈 +0.250%

…asterXML#918) - Allow setNextName(null) / setNextNameIfMissing(null) again (clears name) - Bean serializers set unencoded property name via _setNextPropertyName(); the following writeName() encodes it, so names of properties not written (nulls, filtered out) are no longer encoded - _initWithRootName(): take default namespace from the name actually set, instead of encoding root name a second time - Never encode/decode names in the standard xml: namespace (xml:lang etc), same as XSI names - Rename _nextNameRaw() accessor to _validNextNameRaw() Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions · 2026-10-26T01:44:23Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.89%	📈 +0.200%
Branches	71.28%	📈 +0.310%

…encodes (FasterXML#918) - Never encode names in the xmlns namespace (namespace declarations given as root attributes), same as XSI and xml: names - writeName() no longer carries over XSI/xml/xmlns namespace of previous name to content names (which also skipped their encoding) - Bean property name pending encoding is encoded by writeName() without allocating another QName - startWrappedValue(): reuse encoded wrapper name for items when same - Javadoc: _nextName is unencoded between _setNextPropertyName() and writeName() Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions · 2026-10-26T02:02:22Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.97%	📈 +0.280%
Branches	71.38%	📈 +0.410%

…terXML#918) - Rename package-private _setNextPropertyName() to setNextPropertyName() - startWrappedValue(): accept null wrappedName again (clears name) - _validNextNameRaw(): never consider raw name valid for null _nextName - Factor out _setNextName(raw, encoded); fix _nextIsXsiType javadoc Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions · 2026-10-26T02:13:37Z

🧪 Code Coverage Report

Metric	Coverage	Change
Instructions	77.98%	📈 +0.290%
Branches	71.34%	📈 +0.370%

encode collection wrapper element name via XmlNameProcessor

98f21c1

Sahana2524 force-pushed the wrapper-name-encode branch from 2f112c2 to 98f21c1 Compare September 27, 2026 15:27

cowtowncoder changed the title ~~encode collection wrapper element name via XmlNameProcessor~~ Encode collection wrapper element name via XmlNameProcessor Sep 28, 2026

Merge branch '3.x' into wrapper-name-encode

d729bdb

Merge branch '3.x' into wrapper-name-encode

6cef792

Fix release note placement

ba2c248

Improve tests

3ebb55d

cowtowncoder added the 3.3 label Oct 26, 2026

Merge branch '3.x' into wrapper-name-encode

f5e356f

cowtowncoder changed the title ~~Encode collection wrapper element name via XmlNameProcessor~~ Apply XmlNameProcessor consistently to all written element/attribute names Oct 26, 2026

cowtowncoder approved these changes Oct 26, 2026

View reviewed changes

cowtowncoder merged commit 2c0855d into FasterXML:3.x Oct 26, 2026
3 checks passed

cowtowncoder added this to the 2, milestone Oct 26, 2026

Sahana2524 mentioned this pull request Oct 27, 2026

Enforce maxStringLength/maxNameLength read constraints in XML parser #917

Open

Conversation

Sahana2524 commented Sep 27, 2026 • edited by cowtowncoder Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Sep 27, 2026

🧪 Code Coverage Report

Uh oh!

github-actions Bot commented Sep 27, 2026

🧪 Code Coverage Report

Uh oh!

github-actions Bot commented Sep 28, 2026

🧪 Code Coverage Report

Uh oh!

github-actions Bot commented Oct 6, 2026

🧪 Code Coverage Report

Uh oh!

github-actions Bot commented Oct 6, 2026

🧪 Code Coverage Report

Uh oh!

github-actions Bot commented Oct 6, 2026

🧪 Code Coverage Report

Uh oh!

Sahana2524 commented Oct 23, 2026

Uh oh!

github-actions Bot commented Oct 26, 2026

🧪 Code Coverage Report

Uh oh!

github-actions Bot commented Oct 26, 2026

🧪 Code Coverage Report

Uh oh!

github-actions Bot commented Oct 26, 2026

🧪 Code Coverage Report

Uh oh!

github-actions Bot commented Oct 26, 2026

🧪 Code Coverage Report

Uh oh!

github-actions Bot commented Oct 26, 2026

🧪 Code Coverage Report

Uh oh!

github-actions Bot commented Oct 26, 2026

🧪 Code Coverage Report

Uh oh!

github-actions Bot commented Oct 26, 2026

🧪 Code Coverage Report

Uh oh!

github-actions Bot commented Oct 26, 2026

🧪 Code Coverage Report

Uh oh!

github-actions Bot commented Oct 26, 2026

🧪 Code Coverage Report

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Sahana2524 commented Sep 27, 2026 •

edited by cowtowncoder

Loading