Skip to contentSkip to conversation

tst_net.sh: Declare prefix variable as empty

PATCH New Petr Vorel 3 participants 6 messages Jul 9, 2021 – Jul 12, 2021

lore
Share & tools
lore

Petr Vorel

author
#0 lore raw

PATCH 1/1 tst_net.sh: Declare prefix variable as empty

From: Petr Vorel <[email protected]>

in tst_ipaddr_un(). This is required to fix problem on dash, which (unlike bash and busybox ash implementation) takes value of previously defined variable, i.e. guarding with local does not work:

$ foo=ee; bar() { local foo; echo "foo: '$foo'"; }; bar
foo: 'ee'

It requires declare it as empty:

$ foo=ee; bar() { local foo=; echo "foo: '$foo'"; }; bar
foo: ''

This problem seems to be on various dash versions from 0.5.8 to 0.5.11+git20200708+dd9ef66-5 (on Debian oldstable, stable, testing). It'd require to set all variables, but now fix just prefix, which is used in build.sh and it's quite common name.

---
 testcases/lib/tst_net.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
testcases/lib/tst_net.sh +1 -1
index db1db4790..52975f37b 100644
@@ -408,7 +408,7 @@ tst_ipaddr_un()
 	local max_net_id=$default_max
 	local min_net_id=0
 
-	local counter host_id host_range is_counter max_host_id min_host_id net_id prefix tmp type
+	local counter host_id host_range is_counter max_host_id min_host_id net_id prefix= tmp type
 
 	local OPTIND
 	while getopts "c:h:n:p" opt; do
Signature · 1 line
2.26.2
#1 lore raw

PATCH 1/1 tst_net.sh: Declare prefix variable as empty

Quoted 9 lines · Petr Vorel

From: Petr Vorel <[email protected]>

in tst_ipaddr_un(). This is required to fix problem on dash, which (unlike bash and busybox ash implementation) takes value of previously defined variable, i.e. guarding with local does not work:

$ foo=ee; bar() { local foo; echo "foo: '$foo'"; }; bar
foo: 'ee'

I'm thinking maybe we should avoid defining the same name variables like that. But anyway, your fix should be worked.

Quoted 35 lines

It requires declare it as empty:

$ foo=ee; bar() { local foo=; echo "foo: '$foo'"; }; bar
foo: ''

This problem seems to be on various dash versions from 0.5.8 to 0.5.11+git20200708+dd9ef66-5 (on Debian oldstable, stable, testing). It'd require to set all variables, but now fix just prefix, which is used in build.sh and it's quite common name.

---
 testcases/lib/tst_net.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
testcases/lib/tst_net.sh +0 -1
index db1db4790..52975f37b 100644
@@ -408,7 +408,7 @@ tst_ipaddr_un()
        local max_net_id=$default_max
        local min_net_id=0
 
-       local counter host_id host_range is_counter max_host_id

min_host_id net_id prefix tmp type
+ local counter host_id host_range is_counter max_host_id
min_host_id net_id prefix= tmp type

local OPTIND
while getopts "c:h:n:p" opt; do
--
2.26.2

Signature · 5 lines
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20210712/436494f8/attachment.htm>
author
#2 lore raw

PATCH 1/1 tst_net.sh: Declare prefix variable as empty

Hi Li,

Quoted 8 lines · Petr Vorel
Quoted 6 lines

From: Petr Vorel <[email protected]>

in tst_ipaddr_un(). This is required to fix problem on dash, which (unlike bash and busybox ash implementation) takes value of previously defined variable, i.e. guarding with local does not work:

Quoted 8 lines
Quoted 2 lines
$ foo=ee; bar() { local foo; echo "foo: '$foo'"; }; bar
foo: 'ee'

I'm thinking maybe we should avoid defining the same name variables like that. But anyway, your fix should be worked.

Thanks for a review!
It's a bit strange to add '=' only single variable, maybe I should have rename the variable.

Paranoid approach would be to add '=' to any variable which is assigned only on some circumstance or even any variable which is not assigned to any value. But we probably don't want to do it.

Kind regards,
Petr

#3 lore raw

PATCH 1/1 tst_net.sh: Declare prefix variable as empty

Hi,

Quoted 7 lines · Petr Vorel

Thanks for a review!
It's a bit strange to add '=' only single variable, maybe I should have rename the variable.

Paranoid approach would be to add '=' to any variable which is assigned only on some circumstance or even any variable which is not assigned to any value. But we probably don't want to do it.

I think good common practice would be to add it to all variables, where it matters.
There are probably only very few places, where a local variable is not assigned before being accessed.
Maybe there is some kind of shell code linter, that can find uninitialized variables?
A good linter could also find variables, that should be local, but are not. But I have never looked into shell code linting.

In fs_bind_lib.sh, OPTIND in fs_bind_check probably also requires setting to empty, or better 0.
But maybe dash's implementation of getopts does not require it. I only had to add local OPTIND for busybox sh iirc...

Yet another case of: Shellcode is unpredictable :)

Joerg

author
#4 lore raw

PATCH 1/1 tst_net.sh: Declare prefix variable as empty

Hi Joerg,

Quoted 10 lines

Hi,

Quoted 7 lines · Petr Vorel

Thanks for a review!
It's a bit strange to add '=' only single variable, maybe I should have rename the variable.

Paranoid approach would be to add '=' to any variable which is assigned only on some circumstance or even any variable which is not assigned to any value. But we probably don't want to do it.

Quoted 8 lines

I think good common practice would be to add it to all variables, where it matters.
There are probably only very few places, where a local variable is not assigned before being accessed.
Maybe there is some kind of shell code linter, that can find uninitialized variables?
A good linter could also find variables, that should be local, but are not. But I have never looked into shell code linting.

Thanks for your input.

I'm aware only of shellcheck [1], which I'm not a big fan of (false positives, useless hints). And yes, in this case it warns about 'local' not being POSIX => useless :(.

In testcases/lib/tst_net.sh line 411:

	local counter host_id host_range is_counter max_host_id min_host_id net_id prefix tmp type
        ^-- SC3043: In POSIX sh, 'local' is undefined.
Quoted 2 lines

In fs_bind_lib.sh, OPTIND in fs_bind_check probably also requires setting to empty, or better 0.

We use tst_test.sh in OPTIND=1 ("The index of the next argument to be processed by the getopts builtin command")

Quoted 2 lines

But maybe dash's implementation of getopts does not require it. I only had to add local OPTIND for busybox sh iirc...

I used to set local OPTIND (in tst_net.sh), which is obviously not working on dash.

Quoted 1 line

Yet another case of: Shellcode is unpredictable :)

Yep, I love shell (understand why everything is slowly rewritten into C).

Kind regards,
Petr

Quoted 1 line

Joerg

[1] https://github.com/koalaman/shellcheck

author
#5 lore raw

PATCH 1/1 tst_net.sh: Declare prefix variable as empty

In reply to Joerg Vehlow

Hi all,

Quoted 7 lines

in tst_ipaddr_un(). This is required to fix problem on dash, which (unlike bash and busybox ash implementation) takes value of previously defined variable, i.e. guarding with local does not work:

$ foo=ee; bar() { local foo; echo "foo: '$foo'"; }; bar
foo: 'ee'
Quoted 8 lines

It requires declare it as empty:

$ foo=ee; bar() { local foo=; echo "foo: '$foo'"; }; bar
foo: ''

This problem seems to be on various dash versions from 0.5.8 to 0.5.11+git20200708+dd9ef66-5 (on Debian oldstable, stable, testing). It'd require to set all variables, but now fix just prefix, which is used in build.sh and it's quite common name.

FYI this behavior is expected on dash, quoting man dash(1):
"When a variable is made local, it inherits the initial value and exported and readonly flags from the variable with the same name in the surrounding scope, if there is one."

Kind regards,
Petr