Error parsing "cond && var=value"

[benhoyt]

Error parsing "cond && var=value"

[benhoyt]

Hi @juntuu. FWIW, I've fixed the $++a one in #168, and I'm planning to fix the 1&&x=1 one in #170 (which I don't love, but oh well).

[juntuu]

I haven't gone through all the cases in my repo, but there are three differences (from all gawk, onetrueawk and mawk) I remember having noticed:

1. the $++a one, that has issue open already
2. the one in this issue, also with logical or || and the augmented assignment operators += etc.
3. non-associative binary operators with assignment as the right hand side:
   • comparison operators 1 < x=2, others parse as 1 < (x=2) vs. goawk (1 < x) = 2
   • regex match operators 1 ~ x=2

I'm not sure what would be "the correct" parse for 2. and 3. in terms of the posix spec, but all the other major implementations seem to behave the same way.

Also the 2. and 3. might be somewhat related. At least both parse similarly (using the gawk pretty-printing option):

$ gawk -o- 'BEGIN { 1 < x=2 < y=3 }'
BEGIN {
	1 < (x = 2 < (y = 3))
}

$ gawk -o- 'BEGIN { 1 && x=2 && y=3 }'
BEGIN {
	1 && (x = 2 && (y = 3))
}

Going on a tangent...

I just noticed while writing this, that the handling of the non-associative operators is all over the place when there is no assignment expressions involved:

gawk accepts chaining of match-ops (left-associative), but not comparisons:

$ gawk -o- 'BEGIN { 1 ~ 2 ~ 3 }'
BEGIN {
	(1 ~ 2) ~ 3
}

$ gawk -o- 'BEGIN { 1 < 2 < 3 }'
gawk: cmd. line:1: BEGIN { 1 < 2 < 3 }
gawk: cmd. line:1:               ^ syntax error

onetrueawk rejects both with similar error message:

$ nawk 'BEGIN { 1 ~ 2 ~ 3 }'
nawk: syntax error at source line 1
 context is
	BEGIN { 1 ~ 2 >>>  ~ <<<
nawk: illegal statement at source line 1
nawk: illegal statement at source line 1

and mawk accepts both (left-associative):

$ mawk -W dump 'BEGIN { 1 < 2 < 3 }'
BEGIN
000 .	pushd	1
002 .	pushd	2
004 .	lt
005 .	pushd	3
007 .	lt
008 .	pop
009 .	exit0

In conclusion, I'd hope that any important awk code would be written in more unambiguous style 😅

[benhoyt]

Thanks @juntuu! After I submitted this issue I did go through all the solutions in your repo that were failing in GoAWK, and that's the list I came up with too. Yep, I figured I'd address the cond || var += value and similar cases at the same time -- they're no doubt manifestations of the same issue.

As far as the POSIX spec goes, it says that && and || are higher-precedence than lvalue=expr (and augmented assignment), which is how I'm parsing it. However, it also says that && and || are left-associative and lvalue=expr is right-associative. I actually don't really know what that means! I thought that associativity was only needed when operators have the same precedence. The POSIX spec says, "In expression evaluation, where the grammar is formally ambiguous, higher precedence operators shall be evaluated before lower precedence operators." I'll have to take a closer look and improve my understanding here...