Honestly I had no idea what ctrl+d even did, I just knew it was a convenient way for me to close all the REPL programs I use. The fact that it is similar to pressing enter really surprised me, so I wanted to share this knowledge with you :)
It’s not. You keep insisting that ^D doesn’t send EOF and yet:
$ stty -a | grep eof intr = ^C; quit = ^\; erase = ^?; kill = ^U; eof = ^D; eol = <undef>; $ man stty |grep -A1 eof |head -n2 eof CHAR CHAR will send an end of file (terminate the input)
^D (which is an ASCII EOT character) signals EOF. The thing is that in C every line of a text file must be terminated by a new-line. And so, when you end a file with ^D without a return, you get funky results.
Note: for readers who aren’t aware, the notation
^X
means hold down the ctrl key and type x (without shift).ctrl-a though ctrl-z will send ASCII characters 1 through 26, which are called control characters (because they’re for controling things, and also because you can type them by holding down the control key).
^D is the EOF character.
$ stty -a | grep eof intr = ^C; quit = ^\; erase = ^?; kill = ^U; eof = ^D; eol = <undef>; $ man stty |grep -A1 eof |head -n2 eof CHAR CHAR will send an end of file (terminate the input)
Nope, Chuck Testa: there is no EOF character. Or, one could also say there is an EOF character, but which character it is can be configured on a per-tty basis, and by default it is configured to be
^D
- which (since “D” is the fourth letter of the alphabet) is ASCII character 4, which (as you can see inman ascii
) is called EOT or “end of transmission”.What that
stty
output means is that^D
is the character specified to triggereof
. That means this character is intercepted (by the kernel’s tty driver) and, instead of sending the character to the process reading standard input, the tty “will send an end of file (terminate the input)”.By default
eof
is^D
(EOT), a control character, but it can be set to any character.For instance: run
stty eof x
and now, in that terminal, “x” (by itself, without the control key) will be the EOF character and will behave exactly as^D
did before. (The rest of this comment assumes you are still in a normal default terminal where you have not done that.)But “send an end of file” does not mean sending EOT or any other character to the reading process: as the blog post explains, it actually (counterintuitively) means flushing the buffer - meaning, causing the
read
syscall to return with whatever is in the buffer currently.It is confusing that this functionality is called
eof
, and thestty
man page description of it is even more so, given that it (really!) does actually flush the contents of the buffer toread
- even if the line buffer is not empty, in which case it is not actually indicating end-of-file!You can confirm this is happening by running
cat
and typing a few characters and then hitting^D
, and then typing more, and hitting^D
again. (Each time you flush the buffer,cat
will immediately echo the latest characters that had been buffered, even though you have not hit enter yet.)Or, you can pipe
cat
intopv
and see that^D
also causespv
to receive the buffer contents prior to hitting enter.I guess unix calls this
eof
because this function is most often used to flush an empty buffer, which is how you “send an end of file” to the reader.The empty-
read
-means-EOF semantics are documented, among other places, in the man page for theread()
syscall (man read
):RETURN VALUE On success, the number of bytes read is returned (zero indicates end of file), and the file position is advanced by this number.
If you want to send an actual
^D
(EOT) character through to the process reading standard input, you can escape it using the confusingly-namedlnext
function, which by default is bound to the^V
control character (aka SYN, “synchronous idle”, ASCII character 22):$ man stty|grep lnext -A1 * lnext CHAR CHAR will enter the next character quoted $ stty -a|grep lnext werase = ^W; lnext = ^V; discard = ^O; min = 1; time = 0;
Try it: you can type
echo "
and then ctrl-V and ctrl-D and then"|xxd
(and then enter) and you will see that this is sending ascii character 4.You can also send it with
echo -e '\x04'
. Note that the EOT character does not terminate bash:$ echo -e '\x04\necho see?'|xxd 00000000: 040a 6563 686f 2073 6565 3f0a ..echo see?. $ echo -e '\x04\necho see?'|bash bash: line 1: $'\004': command not found see?
As you can see, it instead interprets it as a command.
(Control characters are perfectly cromulent filenames btw...)
$ echo -e '#!/bin/bash\necho lmao' > ~/.local/bin/$(echo -en '\x04') $ chmod +x ~/.local/bin/$(echo -en '\x04') $ echo -e '\x04\necho see?'|bash lmao see?
Which is why I haven’t wrote ‘EOF character’, ‘EOT’ or ‘EOT character’. Neither have I claimed that
\x4
character is interpreted by the shell as end of file.Edit: Actually, I did say ‘EOF character’ originally (though I still haven’t claimed that it sends EOF character to the program). I’ve updated the comment to clear things up more.
To be more precise, it’s the “EOT” (end of transmission) control character, the 4th symbol in ASCII, from the non-printable character area.
Ctl-D is the End-of-File character. Programs interpret it as “that’s it, the input you were reading has finished”, and react accordingly.
$ cat You sound very nice :) You sound very nice :) Bye<ctl-d>Bye Oh wait, and cool too Oh wait, and cool too <ctl-d> $
The Ctl-D didn’t end the file when i typed “Bye” :( it only worked when I pressed Ctl-D on its own line. So how does cat know that it should ignore the EOF character if there is some text that comes before it?
What Ctl-D does is flush the input to the program, and the program sees how big that input is. If the length of the input is 0 that is interpreted as EOF. So Ctl-D is like Enter because they both flush the input, but Ctl-D is unlike Enter because it does not append a newline before flushing, and as a consequence you can send empty input (aka an EOF “character”) with Ctl-D.
When running cat this way, you are in “cooked mode”. A ctrl-d does nothing on a non-empty line.
The shell usually runs in non-cokked, or raw, mode as well as nonblocking mode. Where it sees (nearly) every key you press as you press them. Which is why it " sees" the ctrl-d even when you are not on an empty line.
You can learn more here:
- https://jvns.ca/blog/2024/11/26/terminal-rules/#rule-3-repls-should-quit-when-you-press-ctrl-d-on-an-empty-line. (And more generally, from her various blog posts on the matter)
- https://www.gnu.org/software/mit-scheme/documentation/stable/mit-scheme-ref/Terminal-Mode.html
A ctrl-d does nothing on a non-empty line.
ctrl-d actually is flushing the buffer regardless of if the line is empty or not.
See my other comment for how you can observe it.
Interesting, I have not heard of these terms before. Thanks for sharing!
I think this adds the bit of nuance that was bugging me: using something like ncurses or vim, presumably when you press a key like ctrl-z or ctrl-d it actually sends the character to the app. It would feel a bit silly if the terminal intercepted the ctrl-d, flushed some buffer, and the program had to reverse engineer whether you pressed ctrl-d or enter or something.
For raw mode, I assume the app asks the tty to please forward some characters to the app. Otherwise, in the default cooked mode, the tty intercepts those control characters to call certain functions. I suppose some REPLs may choose to emulate a cooked mode on top of raw mode, and so they have to handle the \x04 in the same way a tty would to keep it functioning like the user expects. I believe
readline
does something like this, which is why you had to usebash --noediting
for ctrl-d to run the command. Good food for thought :)I also have to say, naming it “cooked mode” is extremely funny as gen z. I love that
This!
It’s merely a buffer flush, in case it’s empty, the program handling the input can choose how to interpret,
cat
decides to do it as an EOF.Reason why it also works as exit.
Ctrl+d terminates input on stdin to your currently running program or shell.
not true. try this:
$ date<C-d>
bash did not terminate stdin, because when i press enter it still runs the command, and my shell continues to work as normal!
you can also try this:
$ bash --noediting $ date<C-d><C-d>
and it will print the date.
so something else is happening here! thats what the link talks about in detail
deleted by creator
For some reason my mobile client didn’t make the article link immediately obvious. That’s actually really interesting. Apparently I was under the same common misconception. So the shell in this case is choosing to continue after detecting the flush.
Ohh I gotcha. Honestly no sweat, its kind of just a bit of fun trivia really :)
Lol wrong again