Note that the peek tests in this PR fail on master, where it calls Base's generic peek(io, T) implementation based on mark and reset. Seems like that indicates a bug somewhere else? I will try to make a minimal reproducer.

Update: see #78

Codecov Report

Patch coverage: 100.00% and project coverage change: +1.46 🎉

Comparison is base (00bf5e8) 85.82% compared to head (763833c) 87.29%.

@@            Coverage Diff             @@
##           master      #77      +/-   ##
==========================================
+ Coverage   85.82%   87.29%   +1.46%     
==========================================
  Files           5        5              
  Lines         381      425      +44     
==========================================
+ Hits          327      371      +44     
  Misses         54       54              

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

CI failure is due to JuliaLang/julia#50532 — need to decide what we want the behavior to be.

Update:: fixed following JuliaLang/julia#50552

For the record, on a simple benchmark

s = randstring("xα∆🐨", 100) * ' '
io = BufferedInputStream(IOBuffer(s));
@btime skipchars(!=(' '), seekstart($io));

this is about 20% faster than the much simpler implementation:

function skipchars(predicate, stream::BufferedInputStream; linecomment=nothing)
    BufferedStreams.checkopen(stream)
    while !eof(stream)
        mark(stream)
        c = read(stream, Char)
        if c === linecomment
            error("unimplemented") # not benchmarking line comments
        elseif predicate(c)
            unmark(stream)
        else
            reset(stream)
            break
        end
    end
    return stream
end

which seems worth it for another 20 lines of code for _readchar(::BufferedInputStream), since fast character-by-character I/O is an important use of buffered input.