[denny0223]

This is an evaluation of upstream #79, whose indented-code fix is valid for the upstream renderer. The fork already preserves the blank line after indented code through 1d8538c and already tests that behavior.

The candidate copies tests/indented_code_test.lua unchanged from upstream a10f025. It contains no runtime changes and is intended to close without merging, after recording the assessment in a closing comment. It is a tests-only extraction, not a full cherry-pick.

The fork handles literal code earlier in the rendering loop and retains heading, thematic-break, and list resets alongside blank-state tracking. Copying upstream's runtime hunk would not add the missing behavior here. Keeping the established fork path avoids unnecessary state-handling churn; keeping the existing regression cases avoids duplicate permanent coverage.

Validation against fork base 94112ca, with Neovim 0.12.0 and 0.12.5:

• Unchanged upstream tests: 2 passed, 0 failed.
• Existing code_isolation_test.lua: 829 passed, 0 failed.
• In an isolated copy, deleting only prev_rendered_blank = code_content:match "^[ \t]*$" ~= nil from the literal indented-code path: 825 passed, 4 expected failures. The failures are output rows and source ownership for the CM117 and wider-list cases; the unchanged candidate still passes all 829 assertions.
• The copied test matches the upstream file byte for byte; StyLua 2.5.2 and Luacheck 1.2.0 pass.

The upstream tests were also run on upstream #79's parent and resulting commit under Neovim 0.12.0: 0/2 passing before, 2/2 passing after. This distinguishes a useful upstream fix from an unnecessary backport for this fork.

[denny0223]

Evaluated and closing without merge. The upstream fix is valid; this fork already implements the behavior and has regression protection, so neither a runtime backport nor duplicate permanent tests are needed.

Sources and scope:

• Upstream: delphinus/md-render.nvim#79, commit a10f025.
• Evaluated fork base: 94112ca; candidate: 87aae89.
• Existing fork implementation: 1d8538c. The literal-code path tracks the actual blank state and also resets heading, thematic-break, and list state. This is the fork's earlier literal-code path, so identical behavior does not require identical code placement.
• This candidate extracts only the unchanged upstream tests. No runtime hunk was imported.

The following commands passed on the candidate with both Neovim 0.12.0 and 0.12.5:

nvim -n -i NONE --headless -u NONE --noplugin -l tests/indented_code_test.lua
# 2 passed, 0 failed
nvim -n -i NONE --headless -u NONE --noplugin -l tests/code_isolation_test.lua
# 829 passed, 0 failed
stylua --check tests/indented_code_test.lua
luacheck --no-cache tests/indented_code_test.lua

To verify that existing coverage detects the regression, an isolated copy removed only prev_rendered_blank = code_content:match "^[ \t]*$" ~= nil from that path. code_isolation_test.lua then reported 825 passed and 4 expected failures on both versions: exact output rows and source ownership for CM117 and the wider-list case. The unchanged candidate was rerun and still passed 829/829. This mutation is not the same as setting the state to false or removing the fork's HR/list resets.

Separately, upstream's original two assertions failed on the parent of a10f025 and passed on that commit under Neovim 0.12.0. The upstream patch addresses a real bug; the decision here is about whether this fork needs the same change.

Revisit if the literal-code state handling is refactored, the cited regression coverage is removed, or a new failing case shows that the fork no longer preserves this behavior. No changes from this evaluation PR enter main.