printstat and awkprintf are very clear: print statement errors
are fatal.
In Jan 2020 [1], to prevent fatal print errors from masquerading
as fclose warnings, every WARNING in closefile and closeall became
FATAL. This broke awk's close and getline functions.
close no longer returns if there's an error, unless the stream
doesn't exist.
getline read errors still return -1, but they are no longer
ignorable. Eventually, one of the closing functions will inspect the
stream with ferror and call FATAL.
In Jul 2020 [2], fatal stdout write errors which had been detectable by
closefile for a few months became invisible, a consequence of switching
standard streams from fclose (which reports flush errors) to freopen
(which ignores them). The Jan 2020 changes which broke getline and
close were themselves partially broken.
The solution is to finish printing before closing. That is to flush
and ferror every stream opened for writing before calling fclose,
pclose, or freopen. A failure to write print statement data is
fatal. A failure to close a flushed stream is a warning. They must
be handled separately.
Every redirected print statement is finished in printstat or awkprintf.
The same is not true of unredirected print statements. To finish
these, stdout must be flushed at some point after the final such
statement. Any problem with that flush is fatal.
Though only stdout needs it, let's defensively finish every stream
opened for writing, so this bug won't recur if someone changes how
redirected streams are flushed.
Write errors on stderr by the implementation are never fatal. When
closing, we only warn of them. Write errors from an application
attempting a redirected print to /dev/stderr are as immediately fatal
as every other redirected print statement.
[1] fed1a562c3
[2] b82b649aa6
IEEE Std 1003.1-2008 mandates that -F str be treated the same as -v
FS=str. For a null string, this was not the case. Since awk(1) documents
that a null string for FS has a specific behavior, make -F '' behave
consistently with -v FS="".
PR:
upstream issue: https://github.com/onetrueawk/awk/issues/127
Sponsored by: Netflix
I botched readrec's definition of a record, when I implemented
RS regular expression support. This is the relevant hunk from the
old diff:
```
- return c == EOF && rr == buf ? 0 : 1;
+ isrec = *buf || !feof(inf);
+ dprintf( ("readrec saw <%s>, returns %d\n", buf, isrec) );
+ return isrec;
```
Problem #1
Unlike testing with EOF, `*buf || !feof(inf)` is blind to stdio
errors. This can cause an infinite loop whose each iteration fabricates
an empty record.
The following demonstration uses standard terminal access control
policy to produce a persistent error condition. Note that the "i/o
error" message does not come from readrec(). It's produced much later
by closeall() at shutdown.
```
$ trap '' SIGTTIN && awk 'END {print NR}' &
[1] 33517
$ # After fg, type ^D
$ fg
trap '' SIGTTIN && awk 'END {print NR}'
13847376
awk: i/o error occurred on /dev/stdin
input record number 13847376, file
source line number 1
```
Each time awk tries to read the terminal from the background,
while ignoring SIGTTIN, the read fails with EIO, getc returns EOF,
the stream's end-of-file indicator remains clear, and `!feof`
erroneously promotoes the empty buffer to an empty record. So long
as the error persists, the stream's position does not advance and
end-of-file is never set.
Problem #2:
When RS is a regex, `*buf || !feof(inf)` can't see an empty record's
terminator at the end of a stream.
```
$ echo a | awk 1 RS='a\n'
$
```
That pipeline should have found one empty record and printed a blank
line, but `*buf || !feof(inf)` considers reaching the end of the
stream the conclusion of a fruitless search. That's only correct when
the terminator is a single character, because a regex RS search can
set the end-of-file marker even when it succeeds.
The Fix
`isrec` must be 0 **iff** no record is found. The correct definition
of "no record" is a failure to find a record terminator and a
failure to find any data (possibly from a final, unterminated
record). Conceptually, for any RS:
```
isrec = (noTERM && noDATA) ? 0 : 1
```
noDATA is an expression that's true if `buf` is empty, false otherwise.
When RS is null or a single character, noTERM is an expression
that is true when the sought after character is not found, false
otherwise. Since the search for a single character can only end with
that character or EOF, noTERM is `c == EOF`.
```
isrec = (c == EOF && rr == buf) ? 0 : 1
```
When RS is a regular expression: noTERM is an expression that is
true if a match for RS is not found, false otherwise. This is simply
the inverse of the result of the function that conducts the search,
`!found`.
```
isrec = (found == 0 && *buf == '\0') ? 0 : 1
```
RS ^-anchoring needs to know if it's reading the first record of a file.
Unfortunately, innew, the flag that the main i/o loop uses to track
this, didn't make it from NetBSD unscathed. This commit restores the
last of the wayward lines.
Without this fix, when reading the first record of an input file named
on the command line, the regular expression machinery will be
misconfigured, precluding a successful match.
Relevant commits:
1. 643a5a3dad (Initial import)
2. ffee7780fe (Restoring innew)
This resulted in the NUL terminator being written to the end of the
buffer which was not the same as the end of the string. That in
turn caused garbage bytes from malloc() to be processed. Also
change the NUL termination to be less error prone by writing the
NUL immediately after the last byte copied.
Reproducible with the following under valgrind:
echo '#!/usr/bin/awk' | awk \
'/^#! ?\/.*\/[a-z]{0,2}awk/ {sub(/^#! ?\/.*\/[a-z]{0,2}awk/,"#! awk"); print}'
If awk prints an error message while when compile_time is still set
to ERROR_PRINTING, don't try to print the context since there is
none. This can happen due to a problem with, e.g., unknown command
line options.
When awk reaches EOF parsing the program file, curpfile is incremented.
However, cursource() uses curpfile without checking it against npfile
which can cause an out of bounds access of pfile[] if there is a syntax
error at the end of the program file.
* Cast to uschar when storing a char in an int that will be used as an index.
Fixes a heap underflow when the input char has the high bit set and
FS is a regex.
* Add regress test for underflow when RS is a regex and input is 8-bit.
* In closeall(), skip stdin and flush std{err,out} instead of closing.
Otherwise awk could fclose(stdin) twice (it may appear more than once)
and closing stderr means awk cannot report errors with other streams.
For example, "awk 'BEGIN { getline < "-" }' < /dev/null" will call
fclose(stdin) twice, with undefined results.
* If closefile() is called on std{in,out,err}, freopen() /dev/null instead.
Otherwise, awk will continue trying to perform I/O on a closed stdio
stream, the behavior of which is undefined.
