tr: Stream output instead of buffering

[BenWiederhake]

This should lower memory consumption, and fixes OOM in some scenarios.

Before, buggy behavior:

$ yes | tr -ds a b | head -n5  # GNU
y
y
y
y
y
$ yes | cargo run -- -ds a b | head -n5  # WARNING: THIS CAUSES OOM!
^C
[$? = 130]
$ yes yes | tr -s ye s | head -n5  # GNU
s
s
s
s
s
$ yes yes | cargo run -- -s ye s | head -n5  # WARNING: THIS CAUSES OOM!
^C
[$? = 130]

After:

$ yes | cargo run -- -ds a b | head -n5
y
y
y
y
y
$ yes yes | cargo run -- -s ye s | head -n5
s
s
s
s
s
$

It might be tempting to factor out the common line translate_input(&mut locked_stdin, &mut buffered_stdout, op); from all if-branches. However, note that op has distinct types in each case, and virtualizing it (e.g. let op: Box<dyn SymbolTranslator>) would incur additional runtime costs. That cost seems unnecessary just to gain a little bit of beauty.

I'm not really sure how to test this. Ideally we could pipe in some data, and NOT close the stdin pipe, and see whether the program produces any partial output. However, I didn't see any methods/functions for doing so, and it sounds like it's too hard to get right to be worth it.

Maybe I could write a BENCHMARKING.md file, like for shuf? What do you suggest?

[github-actions]

GNU testsuite comparison:

Skipping an intermittent issue tests/tail/inotify-dir-recreate (passes in this run but fails in the 'main' branch)

[BenWiederhake]

Changes since last push:

• Nothing, I just want to re-run CI
• CI flaked in test_split::test_round_robin_limited_file_descriptors

Test output of flaky test `test_split::test_round_robin_limited_file_descriptors`

---- test_split::test_round_robin_limited_file_descriptors stdout ----
run: /target/i686-unknown-linux-gnu/debug/coreutils split -n r/40 onehundredlines.txt
thread 'test_split::test_round_robin_limited_file_descriptors' panicked at 'Command was expected to succeed.
stdout = 
 stderr = split: unable to open 'xaz'; aborting
', tests/by-util/test_split.rs:1646:10
stack backtrace:
   0: rust_begin_unwind
             at /rustc/90c541806f23a127002de5b4038be731ba1458ca/library/std/src/panicking.rs:578:5
   1: core::panicking::panic_fmt
             at /rustc/90c541806f23a127002de5b4038be731ba1458ca/library/core/src/panicking.rs:67:14
   2: tests::common::util::CmdResult::success
             at ./tests/common/util.rs:388:9

Sadly, this does not show a backtrace of the failing command. Also, I cannot reproduce this problem. On my machine the command only needs 5 open files at any point in time, so 9 seems like this already includes a lot of wiggle room, so raising this even further is unlikely to fix the issue.

[github-actions]

GNU testsuite comparison:

Skipping an intermittent issue tests/tail/inotify-dir-recreate (passes in this run but fails in the 'main' branch)

[BenWiederhake]

Changes since last push:

• Rebased on current master
• Added documentation for SymbolTranslator::chain, and a short explanation that the *_op variables are the translation operations, as requested by @sylvestre

[BenWiederhake]

Changes since last push:

• Apparently, links in the documentation need <…> around them.

[tertsdiepraam]

I like it! Just one suggestion on the link in the comment.

[BenWiederhake]

Changes since last push(es):

• Rebased on current main
• Make documentation prettier. Thanks @tertsdiepraam!
• Learn the hard way that the "commit suggestion" button doesn't create nice fixups.

[github-actions]

GNU testsuite comparison:

Congrats! The gnu test tests/chown/preserve-root is no longer failing!