[bobsingor]

What changed in the harness

EmbedderTest::OpenDocumentHelper() opened any file that FPDFAvail_IsLinearized() accepts with FPDFAvail_GetDocument(), and only the other branch went through embedpdf_test::LoadCustomDocumentForTest(). In layer mode (EPDF_EMBEDDER_LAYER_MODE=1) a test on a linearized fixture therefore ran on a plain document: no copy-up door, no reads-dont-write check, and EPDFLayer_GetPromotedObjectCount() read 0, so it never showed up as a writer.

Now, in layer mode, a linearized file opens through the layer like any other file. Only a test that asks for linearized loading (LinearizeOption::kMustLinearize, i.e. OpenDocumentLinearized()) keeps the FPDFAvail path. Plain mode is unchanged.

Why that rule: the tests that need progressive loading either ask for it (LoadMainCrossRefTable, LinearizedDocument, Hang1055, PasswordAfterGenerateSave) or build their own FPDF_AVAIL (CreateAvail() + SetDocumentFromAvail(): the data-avail tests and CPDFCreatorEmbedderTest.SaveLinearizedInfo). No test opens a linearized file with the default option and then relies on avail() or the hint tables; GetPageCount() and GetFirstPageNum() already run on non-linearized files.

FPDFAvail_IsLinearized() also checks that /L equals the file length, so 12 of the 14 fixtures that start with /Linearized take this path. annotation_stamp_with_ap.pdf and bug_781804.pdf were saved incrementally after linearizing and already opened through the layer.

Tests that moved onto a layer

20 tests on 10 fixtures (the other two linearized fixtures are only opened with kMustLinearize). All pass in layer mode.

New writers, added to testing/embedpdf_layer_mode_writers.txt:

Test	Objects
EPDFAnnotEmbedderTest.RefitWrapsAStampAcrobatTurnedUpright	5
EPDFPageLabelsEmbedderTest.NoSectionsRemovesTheLabels	1
EPDFPageLabelsEmbedderTest.RefusesSectionsThatBreakTheRules	1
RichTextParityTest.BorderWidthsFixtureMatchesAcrobat	26
RichTextParityTest.CalloutWidthsFixtureMatchesAcrobat	12

RefusesSectionsThatBreakTheRules writes once, at its end; its refusals return before any mutable access. The RichText parity counts are 2 per annotation (annotation + appearance stream), as in the other parity tests.

Now checked as readers (0 objects): CPDFParserEmbedderTest.Bug757705; seven CPDFSecurityHandlerEmbedderTest tests on encrypted.pdf (BadPassword, NoPassword, UserPassword, OwnerPassword, SetRuntimeOwnerPermissions, CheckPasswordPermissionsDoesNotUnlockOwner, CheckPasswordPermissionsIgnoresCurrentOwnerUnlockState); EPDFAnnotEmbedderTest.AppearanceTransformOfAStampAcrobatTurned; EPDFAnnotEmbedderTest.UnrotatedRenderUndoesATurnDrawnInTheStream; EPDFMeasureEmbedderTest.ReadsAllAcrobatImperialMeasurements and ReadsAllAcrobatMetricMeasurements; EPDFPageLabelsEmbedderTest.ReadsTheSectionsOfAFile; FPDFDocEmbedderTest.GetPageLabels; FPDFSaveEmbedderTest.SaveLinearizedDoc.

CPDFSecurityHandlerEmbedderTest.PasswordAfterGenerateSave writes but asks for linearized loading, so it stays plain; it now says so in a comment.

Layer bugs

None found. One test changed: FPDFSaveEmbedderTest.SaveLinearizedDoc failed only in layer mode, and the plain run was the wrong one. The test saved before the simulated download had fetched the metadata stream, so the save left object 26 out (a free entry) while the catalog still said /Metadata 26 0 R. Its byte expectations (0000000026 65535 f, 7912 bytes) described that dangling save. The test now calls SetWholeFileAvailable() first, as SaveLinearizedInfo does, and checks that object 26 is written. Plain and layer mode now write the same 10982 bytes.

Tests (macOS arm64, against embedpdf/main 538900d)

• run_layer_mode.py over the whole embedder suite (1780 tests): writers 500 on main, 505 here; the per-test object counts differ only in the five tests above. Pass/fail/crash counts are the same as main (1555/40/185, all matched by embedpdf_layer_mode_expectations.txt).
• Embedder suite, plain: 1763 passed, 12 failed, 4 skipped, on both main and this branch: the same 12 Mac-baseline failures (progressive render, render goldens, CaptionAppearanceGolden). The independent save checker passed (49 PDFs).
• Unit suite: 1320/1320 pass. pdfium_unittests has none of the changed files as inputs.
• Pixel suite: 51 failures + 2 surprises, the known Mac baseline. pdfium_test has none of the changed files as inputs, so it is the same binary as main's.

Note for #110

With this change, text_rich_comment_acrobat.pdf (#110's fixture) opens through the layer. Measured on a scratch merge of both: EPDFAnnotEmbedderTest.AcrobatCommentWrittenBackReadsTheSame writes 1 object, AcrobatNoteCommentReadsAsRichText reads clean. Whichever of the two merges second needs EPDFAnnotEmbedderTest.AcrobatCommentWrittenBackReadsTheSame 1 in the writers list.