Re: Undetected fatal errors from redirected print

[Andrew J. Schorr]

On Tue, Nov 23, 2021 at 08:42:49PM -0500, Miguel Pineiro Jr. wrote:
> On Tue, Nov 23, 2021, at 6:24 PM, Andrew J. Schorr wrote:
> > The fclose has to do an implicit fflush, of course, before closing the 
> > file. Are
> > you suggesting we should call fflush prior to fclose and then issue an 
> > error if
> > fflush fails (but not if fclose fails)? 
> 
> Yes, to distinguish a fatal error in the print statement's output from 
> subsequent, unrelated stream closing errors. Perhaps something like do_fflush 
> (which understands when a redirection has been declared nonfatal).

This does not feel like an important distinction to me. I will note
that both mawk and nawk throw an error in this scenario:

sh-5.1$ strace -fo /tmp/trace.txt -P /tmp/answer.txt -e fault=write gawk 'BEGIN 
{print "42" > "/tmp/answer.txt"}'

sh-5.1$ strace -fo /tmp/trace.txt -P /tmp/answer.txt -e fault=write mawk 'BEGIN 
{print "42" > "/tmp/answer.txt"}'
mawk: close failed on file /tmp/answer.txt (Function not implemented)

sh-5.1$ strace -fo /tmp/trace.txt -P /tmp/answer.txt -e fault=write nawk 'BEGIN 
{print "42" > "/tmp/answer.txt"}'
nawk: write error on /tmp/answer.txt
 source line number 1

It looks like nawk is flushing after each write, which seems fairly
suboptimal from a performance perspective:

sh-5.1$ strace -fo /tmp/trace.txt -P /tmp/answer.txt -e fault=write nawk 'BEGIN 
{f = "/tmp/answer.txt"; for (i = 1; i < 10; i++) {print i; print i > f}}'
1
nawk: write error on /tmp/answer.txt
 source line number 1

compared to:

sh-5.1$ strace -fo /tmp/trace.txt -P /tmp/answer.txt -e fault=write gawk 'BEGIN 
{f = "/tmp/answer.txt"; for (i = 1; i < 10; i++) {print i; print i > f}}'
1
2
3
4
5
6
7
8
9

sh-5.1$ strace -fo /tmp/trace.txt -P /tmp/answer.txt -e fault=write mawk 'BEGIN 
{f = "/tmp/answer.txt"; for (i = 1; i < 10; i++) {print i; print i > f}}'
1
2
3
4
5
6
7
8
9
mawk: close failed on file /tmp/answer.txt (Function not implemented)

So do you object to the fact that mawk labels this as a close error
instead of a write error? The mawk behavior is fairly similar to the
gawk --lint behavior:

sh-5.1$ strace -fo /tmp/trace.txt -P /tmp/answer.txt -e fault=write gawk --lint 
'BEGIN {f = "/tmp/answer.txt"; for (i = 1; i < 10; i++) {print i; print i > f}}'
1
2
3
4
5
6
7
8
9
gawk: warning: failure status (-1) on file close of `/tmp/answer.txt': Function 
not implemented
gawk: warning: no explicit close of file `/tmp/answer.txt' provided

Regards,
Andy