fix(routing): normalize literal glob components through new URL() by yury-s · Pull Request #41682 · microsoft/playwright

Yury Semikhatsky (yury-s) · 2026-07-07T18:23:01Z

Summary

Since 1.51 (fix: do not change glob pattern when converting to url #34923), resolveGlobBase tokenizes every URL path component before new URL() resolution, so URL normalization no longer applies to the glob. A glob with an explicit default port (http://localhost:80/**) can never match request.url(), which the browser normalizes to http://localhost/. Same regression for percent-encoded literals and IDN hosts.
Fix: only tokenize components that contain glob metacharacters (*, ?, {, }, \); let literal components round-trip through new URL() as before 1.51. Preserves the [Bug]: Incorrect behavior of '?' in the url glob pattern #34915?-mangling fix.

Open question: whether we actually want globs to normalize this way (e.g. :80 silently dropped) or whether the pre-1.51 behavior was itself accidental. Flagging for discussion before merge.

Fixes#41678

resolveGlobBase tokenized every path component before new URL() resolution, so normalization (default ports, percent-encoding, IDN hosts) no longer applied to the glob. A glob with an explicit default port such as http://localhost:80/** could never match request.url(), which is already normalized to http://localhost/. Only tokenize components that contain glob metacharacters; let literal components round-trip through new URL() as they did before 1.51. Fixes: microsoft#41678

github-actions · 2026-07-07T18:58:20Z

Test results for "MCP"

7709 passed, 1235 skipped

Merge workflow run.

github-actions · 2026-07-07T19:05:02Z

Test results for "tests 1"

4 flaky

⚠️ [chromium-library] › library/video.spec.ts:680 › screencast › should capture full viewport on hidpi `@chromium-ubuntu-22.04-node24`
⚠️ [chromium-library] › library/browsercontext-page-event.spec.ts:173 › should work with Ctrl-clicking `@chromium-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/heap.spec.ts:223 › should not leak workers `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/inspector/cli-codegen-3.spec.ts:224 › cli codegen › should generate frame locators (4) `@firefox-ubuntu-22.04-node20`

49500 passed, 1164 skipped

Merge workflow run.

Yury Semikhatsky (yury-s) · 2026-07-07T18:23:01Z

Summary

Since 1.51 (fix: do not change glob pattern when converting to url #34923), resolveGlobBase tokenizes every URL path component before new URL() resolution, so URL normalization no longer applies to the glob. A glob with an explicit default port (http://localhost:80/**) can never match request.url(), which the browser normalizes to http://localhost/. Same regression for percent-encoded literals and IDN hosts.
Fix: only tokenize components that contain glob metacharacters (*, ?, {, }, \); let literal components round-trip through new URL() as before 1.51. Preserves the [Bug]: Incorrect behavior of '?' in the url glob pattern #34915?-mangling fix.

Open question: whether we actually want globs to normalize this way (e.g. :80 silently dropped) or whether the pre-1.51 behavior was itself accidental. Flagging for discussion before merge.

Fixes#41678

resolveGlobBase tokenized every path component before new URL() resolution, so normalization (default ports, percent-encoding, IDN hosts) no longer applied to the glob. A glob with an explicit default port such as http://localhost:80/** could never match request.url(), which is already normalized to http://localhost/. Only tokenize components that contain glob metacharacters; let literal components round-trip through new URL() as they did before 1.51. Fixes: microsoft#41678

github-actions · 2026-07-07T18:58:20Z

Test results for "MCP"

7709 passed, 1235 skipped

Merge workflow run.

github-actions · 2026-07-07T19:05:02Z

Test results for "tests 1"

4 flaky

⚠️ [chromium-library] › library/video.spec.ts:680 › screencast › should capture full viewport on hidpi `@chromium-ubuntu-22.04-node24`
⚠️ [chromium-library] › library/browsercontext-page-event.spec.ts:173 › should work with Ctrl-clicking `@chromium-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/heap.spec.ts:223 › should not leak workers `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/inspector/cli-codegen-3.spec.ts:224 › cli codegen › should generate frame locators (4) `@firefox-ubuntu-22.04-node20`

49500 passed, 1164 skipped

Merge workflow run.

Yury Semikhatsky (yury-s) · 2026-07-07T18:23:01Z

Summary

Since 1.51 (fix: do not change glob pattern when converting to url #34923), resolveGlobBase tokenizes every URL path component before new URL() resolution, so URL normalization no longer applies to the glob. A glob with an explicit default port (http://localhost:80/**) can never match request.url(), which the browser normalizes to http://localhost/. Same regression for percent-encoded literals and IDN hosts.
Fix: only tokenize components that contain glob metacharacters (*, ?, {, }, \); let literal components round-trip through new URL() as before 1.51. Preserves the [Bug]: Incorrect behavior of '?' in the url glob pattern #34915?-mangling fix.

Open question: whether we actually want globs to normalize this way (e.g. :80 silently dropped) or whether the pre-1.51 behavior was itself accidental. Flagging for discussion before merge.

Fixes#41678

resolveGlobBase tokenized every path component before new URL() resolution, so normalization (default ports, percent-encoding, IDN hosts) no longer applied to the glob. A glob with an explicit default port such as http://localhost:80/** could never match request.url(), which is already normalized to http://localhost/. Only tokenize components that contain glob metacharacters; let literal components round-trip through new URL() as they did before 1.51. Fixes: microsoft#41678

github-actions · 2026-07-07T18:58:20Z

Test results for "MCP"

7709 passed, 1235 skipped

Merge workflow run.

github-actions · 2026-07-07T19:05:02Z

Test results for "tests 1"

4 flaky

⚠️ [chromium-library] › library/video.spec.ts:680 › screencast › should capture full viewport on hidpi `@chromium-ubuntu-22.04-node24`
⚠️ [chromium-library] › library/browsercontext-page-event.spec.ts:173 › should work with Ctrl-clicking `@chromium-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/heap.spec.ts:223 › should not leak workers `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/inspector/cli-codegen-3.spec.ts:224 › cli codegen › should generate frame locators (4) `@firefox-ubuntu-22.04-node20`

49500 passed, 1164 skipped

Merge workflow run.

Yury Semikhatsky (yury-s) · 2026-07-07T18:23:01Z

Summary

Since 1.51 (fix: do not change glob pattern when converting to url #34923), resolveGlobBase tokenizes every URL path component before new URL() resolution, so URL normalization no longer applies to the glob. A glob with an explicit default port (http://localhost:80/**) can never match request.url(), which the browser normalizes to http://localhost/. Same regression for percent-encoded literals and IDN hosts.
Fix: only tokenize components that contain glob metacharacters (*, ?, {, }, \); let literal components round-trip through new URL() as before 1.51. Preserves the [Bug]: Incorrect behavior of '?' in the url glob pattern #34915?-mangling fix.

Open question: whether we actually want globs to normalize this way (e.g. :80 silently dropped) or whether the pre-1.51 behavior was itself accidental. Flagging for discussion before merge.

Fixes#41678

resolveGlobBase tokenized every path component before new URL() resolution, so normalization (default ports, percent-encoding, IDN hosts) no longer applied to the glob. A glob with an explicit default port such as http://localhost:80/** could never match request.url(), which is already normalized to http://localhost/. Only tokenize components that contain glob metacharacters; let literal components round-trip through new URL() as they did before 1.51. Fixes: microsoft#41678

github-actions · 2026-07-07T18:58:20Z

Test results for "MCP"

7709 passed, 1235 skipped

Merge workflow run.

github-actions · 2026-07-07T19:05:02Z

Test results for "tests 1"

4 flaky

⚠️ [chromium-library] › library/video.spec.ts:680 › screencast › should capture full viewport on hidpi `@chromium-ubuntu-22.04-node24`
⚠️ [chromium-library] › library/browsercontext-page-event.spec.ts:173 › should work with Ctrl-clicking `@chromium-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/heap.spec.ts:223 › should not leak workers `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/inspector/cli-codegen-3.spec.ts:224 › cli codegen › should generate frame locators (4) `@firefox-ubuntu-22.04-node20`

49500 passed, 1164 skipped

Merge workflow run.

Yury Semikhatsky (yury-s) · 2026-07-07T18:23:01Z

Summary

Since 1.51 (fix: do not change glob pattern when converting to url #34923), resolveGlobBase tokenizes every URL path component before new URL() resolution, so URL normalization no longer applies to the glob. A glob with an explicit default port (http://localhost:80/**) can never match request.url(), which the browser normalizes to http://localhost/. Same regression for percent-encoded literals and IDN hosts.
Fix: only tokenize components that contain glob metacharacters (*, ?, {, }, \); let literal components round-trip through new URL() as before 1.51. Preserves the [Bug]: Incorrect behavior of '?' in the url glob pattern #34915?-mangling fix.

Open question: whether we actually want globs to normalize this way (e.g. :80 silently dropped) or whether the pre-1.51 behavior was itself accidental. Flagging for discussion before merge.

Fixes#41678

resolveGlobBase tokenized every path component before new URL() resolution, so normalization (default ports, percent-encoding, IDN hosts) no longer applied to the glob. A glob with an explicit default port such as http://localhost:80/** could never match request.url(), which is already normalized to http://localhost/. Only tokenize components that contain glob metacharacters; let literal components round-trip through new URL() as they did before 1.51. Fixes: microsoft#41678

github-actions · 2026-07-07T18:58:20Z

Test results for "MCP"

7709 passed, 1235 skipped

Merge workflow run.

github-actions · 2026-07-07T19:05:02Z

Test results for "tests 1"

4 flaky

⚠️ [chromium-library] › library/video.spec.ts:680 › screencast › should capture full viewport on hidpi `@chromium-ubuntu-22.04-node24`
⚠️ [chromium-library] › library/browsercontext-page-event.spec.ts:173 › should work with Ctrl-clicking `@chromium-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/heap.spec.ts:223 › should not leak workers `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/inspector/cli-codegen-3.spec.ts:224 › cli codegen › should generate frame locators (4) `@firefox-ubuntu-22.04-node20`

49500 passed, 1164 skipped

Merge workflow run.

Yury Semikhatsky (yury-s) · 2026-07-07T18:23:01Z

Summary

Since 1.51 (fix: do not change glob pattern when converting to url #34923), resolveGlobBase tokenizes every URL path component before new URL() resolution, so URL normalization no longer applies to the glob. A glob with an explicit default port (http://localhost:80/**) can never match request.url(), which the browser normalizes to http://localhost/. Same regression for percent-encoded literals and IDN hosts.
Fix: only tokenize components that contain glob metacharacters (*, ?, {, }, \); let literal components round-trip through new URL() as before 1.51. Preserves the [Bug]: Incorrect behavior of '?' in the url glob pattern #34915?-mangling fix.

Open question: whether we actually want globs to normalize this way (e.g. :80 silently dropped) or whether the pre-1.51 behavior was itself accidental. Flagging for discussion before merge.

Fixes#41678

resolveGlobBase tokenized every path component before new URL() resolution, so normalization (default ports, percent-encoding, IDN hosts) no longer applied to the glob. A glob with an explicit default port such as http://localhost:80/** could never match request.url(), which is already normalized to http://localhost/. Only tokenize components that contain glob metacharacters; let literal components round-trip through new URL() as they did before 1.51. Fixes: microsoft#41678

github-actions · 2026-07-07T18:58:20Z

Test results for "MCP"

7709 passed, 1235 skipped

Merge workflow run.

github-actions · 2026-07-07T19:05:02Z

Test results for "tests 1"

4 flaky

⚠️ [chromium-library] › library/video.spec.ts:680 › screencast › should capture full viewport on hidpi `@chromium-ubuntu-22.04-node24`
⚠️ [chromium-library] › library/browsercontext-page-event.spec.ts:173 › should work with Ctrl-clicking `@chromium-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/heap.spec.ts:223 › should not leak workers `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/inspector/cli-codegen-3.spec.ts:224 › cli codegen › should generate frame locators (4) `@firefox-ubuntu-22.04-node20`

49500 passed, 1164 skipped

Merge workflow run.

Yury Semikhatsky (yury-s) · 2026-07-07T18:23:01Z

Summary

Since 1.51 (fix: do not change glob pattern when converting to url #34923), resolveGlobBase tokenizes every URL path component before new URL() resolution, so URL normalization no longer applies to the glob. A glob with an explicit default port (http://localhost:80/**) can never match request.url(), which the browser normalizes to http://localhost/. Same regression for percent-encoded literals and IDN hosts.
Fix: only tokenize components that contain glob metacharacters (*, ?, {, }, \); let literal components round-trip through new URL() as before 1.51. Preserves the [Bug]: Incorrect behavior of '?' in the url glob pattern #34915?-mangling fix.

Open question: whether we actually want globs to normalize this way (e.g. :80 silently dropped) or whether the pre-1.51 behavior was itself accidental. Flagging for discussion before merge.

Fixes#41678

resolveGlobBase tokenized every path component before new URL() resolution, so normalization (default ports, percent-encoding, IDN hosts) no longer applied to the glob. A glob with an explicit default port such as http://localhost:80/** could never match request.url(), which is already normalized to http://localhost/. Only tokenize components that contain glob metacharacters; let literal components round-trip through new URL() as they did before 1.51. Fixes: microsoft#41678

github-actions · 2026-07-07T18:58:20Z

Test results for "MCP"

7709 passed, 1235 skipped

Merge workflow run.

github-actions · 2026-07-07T19:05:02Z

Test results for "tests 1"

4 flaky

⚠️ [chromium-library] › library/video.spec.ts:680 › screencast › should capture full viewport on hidpi `@chromium-ubuntu-22.04-node24`
⚠️ [chromium-library] › library/browsercontext-page-event.spec.ts:173 › should work with Ctrl-clicking `@chromium-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/heap.spec.ts:223 › should not leak workers `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/inspector/cli-codegen-3.spec.ts:224 › cli codegen › should generate frame locators (4) `@firefox-ubuntu-22.04-node20`

49500 passed, 1164 skipped

Merge workflow run.

Yury Semikhatsky (yury-s) · 2026-07-07T18:23:01Z

Summary

Since 1.51 (fix: do not change glob pattern when converting to url #34923), resolveGlobBase tokenizes every URL path component before new URL() resolution, so URL normalization no longer applies to the glob. A glob with an explicit default port (http://localhost:80/**) can never match request.url(), which the browser normalizes to http://localhost/. Same regression for percent-encoded literals and IDN hosts.
Fix: only tokenize components that contain glob metacharacters (*, ?, {, }, \); let literal components round-trip through new URL() as before 1.51. Preserves the [Bug]: Incorrect behavior of '?' in the url glob pattern #34915?-mangling fix.

Open question: whether we actually want globs to normalize this way (e.g. :80 silently dropped) or whether the pre-1.51 behavior was itself accidental. Flagging for discussion before merge.

Fixes#41678

resolveGlobBase tokenized every path component before new URL() resolution, so normalization (default ports, percent-encoding, IDN hosts) no longer applied to the glob. A glob with an explicit default port such as http://localhost:80/** could never match request.url(), which is already normalized to http://localhost/. Only tokenize components that contain glob metacharacters; let literal components round-trip through new URL() as they did before 1.51. Fixes: microsoft#41678

github-actions · 2026-07-07T18:58:20Z

Test results for "MCP"

7709 passed, 1235 skipped

Merge workflow run.

github-actions · 2026-07-07T19:05:02Z

Test results for "tests 1"

4 flaky

⚠️ [chromium-library] › library/video.spec.ts:680 › screencast › should capture full viewport on hidpi `@chromium-ubuntu-22.04-node24`
⚠️ [chromium-library] › library/browsercontext-page-event.spec.ts:173 › should work with Ctrl-clicking `@chromium-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/heap.spec.ts:223 › should not leak workers `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-library] › library/inspector/cli-codegen-3.spec.ts:224 › cli codegen › should generate frame locators (4) `@firefox-ubuntu-22.04-node20`

49500 passed, 1164 skipped

Merge workflow run.

Yury Semikhatsky (yury-s) requested review from Dmitry Gozman (dgozman) and Pavel Feldman (pavelfeldman)July 7, 2026 18:23

Pavel Feldman (pavelfeldman) approved these changes Jul 8, 2026
View reviewed changes

Yury Semikhatsky (yury-s) merged commit 3ea100c into microsoft:mainJul 8, 2026
48 checks passed

Yury Semikhatsky (yury-s) deleted the fix-41678 branch July 8, 2026 18:19

Yury Semikhatsky (yury-s) mentioned this pull request Jul 9, 2026
chore: mark v0.1.16 microsoft/playwright-cli#435
Merged

Yusuke Iwaki (YusukeIwaki) mentioned this pull request Jul 31, 2026
[Ports]: Backport client side changes for 1.62 YusukeIwaki/playwright-ruby-client#389
Closed