Markdown tests suceed/fails depending on how you run them

[KristofferC]

Fail:

❯ ./julia test/runtests.jl Markdown 
...

Test  (Worker) | Time (s) | GC (s) | GC % | Alloc (MB) | RSS (MB)
Markdown   (1) |        started at 2025-12-26T22:09:45.537
Markdown   (1) |         failed at 2025-12-26T22:09:48.215
Test Failed at /Users/kc/julia/usr/share/julia/stdlib/v1.14/Markdown/test/runtests.jl:1185
  Expression: String(take!(buf)) in ("  \e[3memph\e[23m", "  \e[4memph\e[24m")
   Evaluated: "  \e[22m\e[3memph\e[39m\e[49m\e[22m\e[23m" in ("  \e[3memph\e[23m", "  \e[4memph\e[24m")
Test Failed at /Users/kc/julia/usr/share/julia/stdlib/v1.14/Markdown/test/runtests.jl:1195
  Expression: startswith(lines[var"begin" + 1], Regex(' ' ^ Markdown.margin * "\e\\[[34]m"))
   Evaluated: startswith("  \e[22m\e[3mMarkdown\e[39m\e[49m\e[22m\e[23m", r"  [[34]m")
Test Failed at /Users/kc/julia/usr/share/julia/stdlib/v1.14/Markdown/test/runtests.jl:1204
  Expression: all((l->begin
            #= /Users/kc/julia/usr/share/julia/stdlib/v1.14/Markdown/test/runtests.jl:1204 =#
            startswith(l, ' ' ^ Markdown.margin * StyledStrings.ANSI_STYLE_CODES.bold_weight) || startswith(l, StyledStrings.ANSI_STYLE_CODES.bold_weight * ' ' ^ Markdown.margin)
        end), lines)
   Evaluated: all(Main.Test2Main_Markdown.var"#3#4"(), SubString{String}["\e[38;2;130;170;255m\e[1m  \e[22mMarkdownMarkdownMarkdown\e[39m\e[49m\e[22m\e[22m\e[22m", "  \e[38;2;130;170;255m\e[22m\e[1m≡≡≡≡≡≡≡≡\e[39m\e[49m\e[22m"])


Test Summary: | Pass  Fail  Total  Time
  Overall     |  270     3    273  3.8s
    Markdown  |  270     3    273  3.4s
    FAILURE

Success:

❯ make test-Markdown                
...

Test  (Worker) | Time (s) | GC (s) | GC % | Alloc (MB) | RSS (MB)
Markdown   (1) |        started at 2025-12-26T22:10:48.639
Markdown   (1) |     2.63 |   0.08 |  3.0 |     414.57 |   448.45

Test Summary: | Pass  Total  Time
  Overall     |  273    273  3.9s
    SUCCESS

Some StyledString stuff?

[vishal2005025]

I’ve opened PR #60538 that normalizes trailing \e[0m ANSI reset codes in the affected Markdown tests. This removes the environment-dependent behavior while preserving the original test assertions and rendering semantics. please let me know if any changes are required.

[fingolfin]

Interestingly, for me your first input, ./julia test/runtests.jl Markdown, works fine. This is on macOS 15.7.3 on an M1 Mac. I wonder if this is related to e.g. the Terminal emulator we use?

[fingolfin]

When I ssh into a Gentoo server and try it there, it also works. So this makes me suspect terminfo or the terminal emulator or something like that? Unfortunately I am not really an expert on that myself.