175568 2017-08-15 03:50:37 -0700 data: URL base64 handling different from atob() 2023-09-13 13:09:19 -0700 1 1 1 Unclassified WebKit DOM WebKit Nightly Build Unspecified Unspecified RESOLVED FIXED https://bugs.webkit.org/show_bug.cgi?id=148128 https://bugs.webkit.org/show_bug.cgi?id=261524 InRadar P2 Normal --- 255442 1 annevk annevk achristensen ap beidson cdumez dbates d koivisto mathias rbuis webkit-bug-importer oldest_to_newest 1338653 0 annevk 2017-08-15 03:50:37 -0700 Tests: https://github.com/w3c/web-platform-tests/pull/6890 Eventual standard: https://github.com/whatwg/fetch/pull/579 It seems better if data: URLs and atob() are aligned. 1338842 1 ap 2017-08-15 13:33:19 -0700 Maybe, but don't they explicitly implement different specs (base64 vs. base64url)? See also https://github.com/whatwg/html/issues/351 1339092 2 annevk 2017-08-16 00:05:04 -0700 They don't, but it does seem like that's how Safari implements data: URLs, contrary to other browsers. That's why data:;base64,___ decodes in Safari and not in Chrome and Firefox. I'm surprised this hasn't caused you more compatibility issues. I'll make sure to add some additional tests for this. 1339313 3 cdumez 2017-08-16 15:17:23 -0700 Looking at DataURLDecoder, I see the following logic: // First try base64url. if (!base64URLDecode(task.encodedData.toStringWithoutCopying(), buffer)) { // Didn't work, try unescaping and decoding as base64. auto unescapedString = decodeURLEscapeSequences(task.encodedData.toStringWithoutCopying()); if (!base64Decode(unescapedString, buffer, Base64IgnoreSpacesAndNewLines)) return; } 1339316 4 cdumez 2017-08-16 15:23:56 -0700 (In reply to Chris Dumez from comment #3) > Looking at DataURLDecoder, I see the following logic: > // First try base64url. > if (!base64URLDecode(task.encodedData.toStringWithoutCopying(), buffer)) > { > // Didn't work, try unescaping and decoding as base64. > auto unescapedString = > decodeURLEscapeSequences(task.encodedData.toStringWithoutCopying()); > if (!base64Decode(unescapedString, buffer, > Base64IgnoreSpacesAndNewLines)) > return; > } This was introduced in Bug 148128 when WebKit started doing data URL decoding by itself. Before this, I believe we let CFNetwork decode those for us (No idea exactly how this behaved). 1339423 5 koivisto 2017-08-16 21:33:43 -0700 It is there to match the earlier CFNetwork behavior. I believe some tests fail if you remove it. 1948457 6 annevk 2023-04-13 00:24:54 -0700 Rob fixed the tests in https://github.com/WebKit/WebKit/commit/7b467a9d964d4214cfa29005135d429279af7d00 it seems, but I don't get the impression the code is shared. 1948458 7 annevk 2023-04-13 00:34:00 -0700 Okay, this in WebLoaderStrategy.cpp seems bad: auto mode = DataURLDecoder::Mode::Legacy; if (request.requester() == ResourceRequestRequester::Fetch) mode = DataURLDecoder::Mode::ForgivingBase64; Removing DataURLDecoder::Mode::Legacy completely and seeing what breaks seems like a logical next step here. 1948459 8 webkit-bug-importer 2023-04-13 01:01:07 -0700 <rdar://problem/107982669> 1948460 9 annevk 2023-04-13 01:02:57 -0700 Pull request: https://github.com/WebKit/WebKit/pull/12693 1948825 10 ews-feeder 2023-04-14 10:54:18 -0700 Committed 262976@main (2518c514c679): <https://commits.webkit.org/262976@main> Reviewed commits have been landed. Closing PR #12693 and removing active labels.