Limit implicit show in REPL to printing 20 KiB by default (#53959)

closes #40735

---------

Co-authored-by: Jameson Nash <vtjnash@gmail.com>

NEWS.md

@@ -167,6 +167,8 @@ Standard library changes
 - the REPL will now warn if it detects a name is being accessed from a module which does not define it (nor has a submodule which defines it),
   and for which the name is not public in that module. For example, `map` is defined in Base, and executing `LinearAlgebra.map`
   in the REPL will now issue a warning the first time occurs. ([#54872])
+- When an object is printed automatically (by being returned in the REPL), its display is now truncated after printing 20 KiB.
+  This does not affect manual calls to `show`, `print`, and so forth. ([#53959])
 
 #### SuiteSparse
 

stdlib/REPL/src/REPL.jl

@@ -484,10 +484,70 @@ function repl_backend_loop(backend::REPLBackend, get_module::Function)
     return nothing
 end
 
+SHOW_MAXIMUM_BYTES::Int = 20480
+
+# Limit printing during REPL display
+mutable struct LimitIO{IO_t <: IO} <: IO
+    io::IO_t
+    maxbytes::Int
+    n::Int # max bytes to write
+end
+LimitIO(io::IO, maxbytes) = LimitIO(io, maxbytes, 0)
+
+struct LimitIOException <: Exception
+    maxbytes::Int
+end
+
+function Base.showerror(io::IO, e::LimitIOException)
+    print(io, "$LimitIOException: aborted printing after attempting to print more than $(Base.format_bytes(e.maxbytes)) within a `LimitIO`.")
+end
+
+function Base.write(io::LimitIO, v::UInt8)
+    io.n > io.maxbytes && throw(LimitIOException(io.maxbytes))
+    n_bytes = write(io.io, v)
+    io.n += n_bytes
+    return n_bytes
+end
+
+# Semantically, we only need to override `Base.write`, but we also
+# override `unsafe_write` for performance.
+function Base.unsafe_write(limiter::LimitIO, p::Ptr{UInt8}, nb::UInt)
+    # already exceeded? throw
+    limiter.n > limiter.maxbytes && throw(LimitIOException(limiter.maxbytes))
+    remaining = limiter.maxbytes - limiter.n # >= 0
+
+    # Not enough bytes left; we will print up to the limit, then throw
+    if remaining < nb
+        if remaining > 0
+            Base.unsafe_write(limiter.io, p, remaining)
+        end
+        throw(LimitIOException(limiter.maxbytes))
+    end
+
+    # We won't hit the limit so we'll write the full `nb` bytes
+    bytes_written = Base.unsafe_write(limiter.io, p, nb)
+    limiter.n += bytes_written
+    return bytes_written
+end
+
 struct REPLDisplay{Repl<:AbstractREPL} <: AbstractDisplay
     repl::Repl
 end
 
+function show_limited(io::IO, mime::MIME, x)
+    try
+        # We wrap in a LimitIO to limit the amount of printing.
+        # We unpack `IOContext`s, since we will pass the properties on the outside.
+        inner = io isa IOContext ? io.io : io
+        wrapped_limiter = IOContext(LimitIO(inner, SHOW_MAXIMUM_BYTES), io)
+        # `show_repl` to allow the hook with special syntax highlighting
+        show_repl(wrapped_limiter, mime, x)
+    catch e
+        e isa LimitIOException || rethrow()
+        printstyled(io, """…[printing stopped after displaying $(Base.format_bytes(e.maxbytes)); call `show(stdout, MIME"text/plain"(), ans)` to print without truncation]"""; color=:light_yellow, bold=true)
+    end
+end
+
 function display(d::REPLDisplay, mime::MIME"text/plain", x)
     x = Ref{Any}(x)
     with_repl_linfo(d.repl) do io
@@ -504,7 +564,7 @@ function display(d::REPLDisplay, mime::MIME"text/plain", x)
             # this can override the :limit property set initially
             io = foldl(IOContext, d.repl.options.iocontext, init=io)
         end
-        show_repl(io, mime, x[])
+        show_limited(io, mime, x[])
         println(io)
     end
     return nothing

stdlib/REPL/test/repl.jl

@@ -1964,6 +1964,46 @@ end
     @test undoc == [:AbstractREPL, :BasicREPL, :LineEditREPL, :StreamREPL]
 end
 
+struct A40735
+    str::String
+end
+
+# https://github.com/JuliaLang/julia/issues/40735
+@testset "Long printing" begin
+    previous = REPL.SHOW_MAXIMUM_BYTES
+    try
+        REPL.SHOW_MAXIMUM_BYTES = 1000
+        str = string(('a':'z')...)^50
+        @test length(str) > 1100
+        # For a raw string, we correctly get the standard abbreviated output
+        output = sprint(REPL.show_limited, MIME"text/plain"(), str; context=:limit => true)
+        hint = """call `show(stdout, MIME"text/plain"(), ans)` to print without truncation"""
+        suffix = "[printing stopped after displaying 1000 bytes; $hint]"
+        @test !endswith(output, suffix)
+        @test contains(output, "bytes ⋯")
+        # For a struct without a custom `show` method, we don't hit the abbreviated
+        # 3-arg show on the inner string, so here we check that the REPL print-limiting
+        # feature is correctly kicking in.
+        a = A40735(str)
+        output = sprint(REPL.show_limited, MIME"text/plain"(), a; context=:limit => true)
+        @test endswith(output, suffix)
+        @test length(output) <= 1200
+        # We also check some extreme cases
+        REPL.SHOW_MAXIMUM_BYTES = 1
+        output = sprint(REPL.show_limited, MIME"text/plain"(), 1)
+        @test output == "1"
+        output = sprint(REPL.show_limited, MIME"text/plain"(), 12)
+        @test output == "1…[printing stopped after displaying 1 byte; $hint]"
+        REPL.SHOW_MAXIMUM_BYTES = 0
+        output = sprint(REPL.show_limited, MIME"text/plain"(), 1)
+        @test output == "…[printing stopped after displaying 0 bytes; $hint]"
+        @test sprint(io -> show(REPL.LimitIO(io, 5), "abc")) == "\"abc\""
+        @test_throws REPL.LimitIOException(1) sprint(io -> show(REPL.LimitIO(io, 1), "abc"))
+    finally
+        REPL.SHOW_MAXIMUM_BYTES = previous
+    end
+end
+
 @testset "Dummy Pkg prompt" begin
     # do this in an empty depot to test default for new users
     withenv("JULIA_DEPOT_PATH" => mktempdir() * (Sys.iswindows() ? ";" : ":"), "JULIA_LOAD_PATH" => nothing) do