[CCCLL]

What is the purpose of the change?

Fixes #16563.

Servlet code running in Triple REST (dubbo-triple-servlet) sent wrong Location headers when it redirected the usual ways:

• ServletHttpRequestAdapter.getContextPath() and DummyServletContext.getContextPath() returned "/", so sendRedirect(getContextPath() + "/login") sent //login. The servlet API returns "" for the root context. They now return "".
• ServletHttpResponseAdapter.encodeURL, encodeRedirectURL, encodeUrl and encodeRedirectUrl returned URLEncoder.encode(url), so sendRedirect(encodeRedirectURL("/login?next=/orders")) sent %2Flogin%3Fnext%3D%2Forders. In the servlet API these methods only add a session id for URL rewriting and otherwise return the URL unchanged. Dubbo does not rewrite URLs for sessions, so they now return the URL as it is.

Both are needed for code like Spring Security's DefaultRedirectStrategy, which calls encodeRedirectURL(getContextPath() + url): with only the second change it would send //login?next=/orders. The PR has one commit for each, the context path first.

Behavior change

• encodeURL, encodeRedirectURL, encodeUrl and encodeRedirectUrl return their argument unchanged instead of a form-encoded copy.
• HttpServletRequest#getContextPath() and ServletContext#getContextPath() return "" instead of "/".
• Code that used encodeURL to percent-escape a value now gets the value back unescaped, as with any servlet container.
• The generated jakarta adapters change the same way, since they are copied from these sources at build time.

Not covered

• A non-root context: when the protocol has a contextpath, or TripleFilter runs in a webapp that is not deployed at the root, getContextPath() now returns "" instead of "/" (a redirect to getContextPath() + "/login" goes to /login instead of //login), but a servlet container would return that context path and leave it out of getServletPath(). Doing that needs the context path of the matched service or of the container request, so I left it out of this PR.

Verification

• New ServletHttpRequestAdapterTest (empty context path for the request and the servlet context) and ServletHttpResponseAdapterTest (the four methods return /login?next=/orders, http://example.com/a b?x=1#y and "" unchanged). They fail before the change (expected: <> but was: </>, expected: </login?next=/orders> but was: <%2Flogin%3Fnext%3D%2Forders>) and pass after it. Putting the old code back into any one of the six methods makes a test fail. The tests use the javax classes because the jakarta copies are only generated on JDK 17+.
• mvn clean test verify of dubbo-triple-servlet and dubbo-rest-spring with the options of the CI unit test job, on JDK 8, 11, 17, 21 and 25: 61 tests, 0 failures each time (11 skipped, the existing @Disabled SpringMvcRestProtocolTest); checkstyle and spotless are clean.
• End to end, the three redirects from the issue, compared with the same servlet code in Tomcat 9.0.121 (Spring Security 5.8.16 for /security):

Request	3.3.6	this PR	Tomcat 9.0.121
/redirect	%2Flogin%3Fnext%3D%2Forders	/login?next=/orders	/login?next=/orders
/context	//login?next=/orders	/login?next=/orders	/login?next=/orders
/security	%2F%2Flogin%3Fnext%3D%2Forders	/login?next=/orders	/login?next=/orders

Checklist

• Make sure there is a GitHub_issue field for the change.
• Write a pull request description that is detailed enough to understand what the pull request does, how, and why.
• Write necessary unit-test to verify your logic correction. If the new feature or significant change is committed, please remember to add sample in dubbo samples project.
• Make sure gitHub actions can pass. Why the workflow is failing and how to fix it?

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 60.98%. Comparing base (1afcdf7) to head (984d823).

Additional details and impacted files

@@             Coverage Diff              @@
##                3.3   #16564      +/-   ##
============================================
+ Coverage     59.02%   60.98%   +1.95%     
  Complexity       15       15              
============================================
  Files          1953     1953              
  Lines         89284    89284              
  Branches      13477    13477              
============================================
+ Hits          52700    54449    +1749     
+ Misses        30877    29242    -1635     
+ Partials       5707     5593     -114     

Flag	Coverage Δ
integration-tests-java21	32.13% <0.00%> (?)
integration-tests-java8	32.20% <0.00%> (?)
samples-tests-java21	32.13% <0.00%> (?)
samples-tests-java8	29.78% <0.00%> (?)
unit-tests-java11	59.21% <100.00%> (+<0.01%)	⬆️
unit-tests-java17	58.69% <100.00%> (+0.03%)	⬆️
unit-tests-java21	58.71% <100.00%> (+0.02%)	⬆️
unit-tests-java25	58.65% <100.00%> (-0.02%)	⬇️
unit-tests-java8	59.24% <100.00%> (-0.03%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[EarthChen]

Verified locally on 3.3 (merge-base 1afcdf7883, PR head 984d823420).

Reproduction. With the three main files reverted to the base, the new tests fail: getContextPath() returns / instead of an empty string, encodeRedirectURL("/login?next=/orders") returns %2Flogin%3Fnext%3D%2Forders, and encodeURL encodes the whole URL. On the PR head both test classes pass (1 + 3 tests), and Spotless and checkstyle pass for the module.

Contract. The fix matches the Servlet API: getContextPath() returns an empty string for the root context, and encodeURL / encodeRedirectURL return the URL unchanged when no session id is added. Dubbo does not rewrite URLs for sessions.

Side effects. I found no reader of these getContextPath() implementations. Routing uses the container's getRequestURI(). RequestUtils.encodeURL is still used by the JAX-RS adapter and tests, and the PR removed the now-unused import.

Minor, non-blocking: the empty-string case in ServletHttpResponseAdapterTest also passes on the old code, and no test covers sendRedirect end to end. A non-root context path is still unsupported, as the issue notes.