What unexpected things can happen if a user runs commands expecting a text file on input lacking a file-final newline?
It is often taught that in Unix/Linux text files should end with newline characters. The reason given (orally) to me by various sources was that "some commands (such as wc) assume or require a newline at the end of a text file". (Other commands I am aware of are: sed, read, cron/crontab, and (in terms of output:) sort.) This led me to ask the following question:
Which commands in Unix/Linux assume that text files end in a newline?
It seems that the proper answer to that question is something like this:
all commands that require a "text file" according to the "INPUT FILES" section of their POSIX specification
This is because POSIX-compliant text files should end in a newline character '\n' (U+0010). (The rationale has been discussed at length on Stack Overflow and is not the subject of this question.)
That is, if I understand correctly, if one only works within Unix & Linux, one only rarely encounters files with textual data that don't end in a newline. Furthermore, the statement that "some commands assume or require a newline at the end of a text file" is misleading, because in fact all commands that manipulate text files have such a requirement, simply because having a file-final newline is a POSIX requirement for text files.
With all that said, I would hereby like to ask:
What are some unexpected things that can happen if a Unix/Linux user forgets to check textual input to commands for file-final newlines?
To elaborate on "unexpected", let <lines> denote a sequence of POSIX lines and <def-line> a defective line (defined as the result of subtracting the final newline character from a non-empty POSIX line). In what follows, def-line thus represents the trailing characters after the input's last newline.
- Let us assume that the expected behavior is
f(<lines><def-line>) = f(<lines><def-line>\n)or
f(<lines><def-line>) ≅ f(<lines><def-line>\n).
Here, approximate equality (≅) covers the case of "the same" newline being missing from the output on the left-hand side. Unexpected behavior would then be
f(<lines><def-line>) ≠ f(<lines><def-line>\n)or
f(<lines><def-line>) ≇ f(<lines><def-line>\n). - A special case of this is
f(<lines><def-line>) = f(<lines>).
The answer needn't be a complete list; a list of common "gotchas" might be enough to teach a new user to be careful to check for text-file-final newlines. Historical gotchas are welcome; I am aware that commands change over time (and also exist in different implementations).
Here are some realistic sources of files that are POSIX-compliant text files, except for the fact that they are missing a file-final newline:
- text files saved with various text editors that don't enforce a file-final newline, such as Sublime Text
- code saved with IDEs that don't enforce a file-final newline
- files with textual data from governments, used for research purposes
- files from other OSes that were transferred in a way that doesn't append missing file-final newlines
Note also that word processors in general don't enforce document-final newlines, so users who are new to handling text files in Unix/Linux might find it surprising that text-file-final newlines are a POSIX requirement. They probably haven't given this issue much thought and will (if questioned) likely guess that a text file is either
- lines separated by newline characters or
- what POSIX defines, except with the last line's final newline character being optional.
It is easy to point users to the standard and state that behavior on malformed input is undefined, but in reality:
- tools don't behave randomly on malformed input (they certainly don't normally erase your hard drive),
- most users learn Unix/Linux by example and by trial-and-error, not by reading standards documents, and
- many tools make allowances for malformed input, because it occurs in the wild – perhaps tools should emit warnings more often, but that is another topic.
I firmly believe that people should be using commands with properly formatted input, and I am not endorsing usage of POSIX commands in a non-POSIX-conformant way. Perhaps the actual issue is that POSIX conventions aren't taught well. However, a list of pitfalls might help casual users of Unix/Linux learn about what could go (or already might have gone) wrong when input isn't properly formatted.
3 answers
You are accessing this answer with a direct link, so it's being shown above all other answers regardless of its score. You can return to the normal view.
I would consider your own tools as the central use case. Seriously, do you want to litter every while read (or your language's equivalent) with this pesky corner case handling?
while read -r line || [ -n "${line-}" ]
do
: something with "$line"
done <"$input_file"
Having a single well-defined behavior also reduces the need for speculation. With malformed files, what exactly should tail -n 1 return? Which line(s) should the $ address in sed (which matches on the final line in well-formed files) match on?
You might argue what you think the "obvious" answer to these questions should be, but since that's non-standard behavior, you can never rely on these constructs to work portably, and so you are needlessly restricting the usefulness of your scripts.
As noted elsewhere, cat and wc -l on malformed files are another two use cases which commonly fail; but the vast bulk of broken tools will probably be local or informal tools which were never tested for this particular corner case.
0 comment threads
I tried to think of some for a while, but couldn't find any good ones. That said, there's plenty of programs I don't know. For reference:
$ echo -en "hello\nworld" | tee 1.txt | bat -A
───────┬─────────────────────────────────────────
│ STDIN
│ Size: -
───────┼─────────────────────────────────────────
1 │ hello␊
2 │ world
───────┴─────────────────────────────────────────
$ echo -e "hello\nworld" | tee 2.txt | bat -A
───────┬─────────────────────────────────────────
│ STDIN
│ Size: -
───────┼─────────────────────────────────────────
1 │ hello␊
2 │ world␊
───────┴─────────────────────────────────────────
-
wcappears to count newlines, sowc 1.txtincorrectly returns 1 for line count (you've mentioned this already) -
sortanduniqbehave sensibly - Python
readlines()returns a list where the last item doesn't have a line ending, but this wouldn't normally break anything since usually people callstrip()on each line (or setkeepends=False)
It's hard to think of an example which is not contrived.
From a programmer's perspective, when your program has some specific requirement for input, you would want to either detect it and fail, or automatically correct it. Both are quite easy for missing terminal newlines. Standards are nice, but when you have a few thousand users or more, you simply cannot assume that not a single one would flout them. I would be surprised if there are any real cases where this is a problem, but I'm curious to see what others will post.
I do know one common case: It can mess up your shell prompt to print such files:
$ bash
$ cat 1.txt
hello
world$
Of course, this is easily resolved by pressing Enter to get a new prompt. Note also how I had to open bash for this - my normal shell, fish, handles it:
$ cat 1.txt
hello
world⏎
$
As does zsh:
$ cat 1.txt
hello
world%
$
And of course, cat behaves a bit surprisingly (well, not really):
$ cat * | bat -A
───────┬─────────────────────────────────────────
│ STDIN
│ Size: -
───────┼─────────────────────────────────────────
1 │ hello␊
2 │ worldhello␊
3 │ world␊
───────┴─────────────────────────────────────────
1 comment thread
I’ve had some tools drop the last incomplete line on some OSes.
For example:
$ printf 'foo\nbar\nbaz' | sed 's/x/y/'
foo
bar
$ _
I don’t recall which systems exactly these were, but I know I ran into it multiple times over the years, and POSIX allows this (I checked, it’s the application’s/user’s duty to provide the final newline).
0 comment threads
3 comment threads