[PATCH] exec: Remove the duplicated check in parse_cpu_option()

[PATCH] exec: Remove the duplicated check in parse_cpu_option()

[Gavin Shan]

The @cpu_option shouldn't be NULL, otherwise assertion from g_strsplit()
should be raised as below message indicates. So it's meaningless to validate
@model_pices[0] in parse_cpu_option() as it shouldn't be NULL either.

   qemu-system-aarch64: GLib: g_strsplit: assertion 'string != NULL' failed

This just removes the check and unused message.

Signed-off-by: Gavin Shan <gshan@redhat.com>
---
 exec.c | 5 -----
 1 file changed, 5 deletions(-)

diff --git a/exec.c b/exec.c
index ffdb518535..3cff459e43 100644
--- a/exec.c
+++ b/exec.c
@@ -963,11 +963,6 @@ const char *parse_cpu_option(const char *cpu_option)
     const char *cpu_type;
 
     model_pieces = g_strsplit(cpu_option, ",", 2);
-    if (!model_pieces[0]) {
-        error_report("-cpu option cannot be empty");
-        exit(1);
-    }
-
     oc = cpu_class_by_name(CPU_RESOLVING_TYPE, model_pieces[0]);
     if (oc == NULL) {
         error_report("unable to find CPU model '%s'", model_pieces[0]);
-- 
2.23.0

Re: [PATCH] exec: Remove the duplicated check in parse_cpu_option()

[Greg Kurz]

On Fri,  6 Dec 2019 17:33:37 +1100
Gavin Shan <gshan@redhat.com> wrote:

> The @cpu_option shouldn't be NULL, otherwise assertion from g_strsplit()
> should be raised as below message indicates. So it's meaningless to validate
> @model_pices[0] in parse_cpu_option() as it shouldn't be NULL either.
> 
>    qemu-system-aarch64: GLib: g_strsplit: assertion 'string != NULL' failed
> 
> This just removes the check and unused message.
> 

Hrm... the check isn't about @cpu_option being NULL. It is about filtering out
invalid syntaxes like:

-cpu ''

or

-cpu ,some-prop

> Signed-off-by: Gavin Shan <gshan@redhat.com>
> ---
>  exec.c | 5 -----
>  1 file changed, 5 deletions(-)
> 
> diff --git a/exec.c b/exec.c
> index ffdb518535..3cff459e43 100644
> --- a/exec.c
> +++ b/exec.c
> @@ -963,11 +963,6 @@ const char *parse_cpu_option(const char *cpu_option)
>      const char *cpu_type;
>  
>      model_pieces = g_strsplit(cpu_option, ",", 2);
> -    if (!model_pieces[0]) {
> -        error_report("-cpu option cannot be empty");
> -        exit(1);
> -    }
> -
>      oc = cpu_class_by_name(CPU_RESOLVING_TYPE, model_pieces[0]);
>      if (oc == NULL) {
>          error_report("unable to find CPU model '%s'", model_pieces[0]);

Re: [PATCH] exec: Remove the duplicated check in parse_cpu_option()

[Gavin Shan]

On 12/7/19 3:58 AM, Greg Kurz wrote:
> On Fri,  6 Dec 2019 17:33:37 +1100
> Gavin Shan <gshan@redhat.com> wrote:
> 
>> The @cpu_option shouldn't be NULL, otherwise assertion from g_strsplit()
>> should be raised as below message indicates. So it's meaningless to validate
>> @model_pices[0] in parse_cpu_option() as it shouldn't be NULL either.
>>
>>     qemu-system-aarch64: GLib: g_strsplit: assertion 'string != NULL' failed
>>
>> This just removes the check and unused message.
>>
> 
> Hrm... the check isn't about @cpu_option being NULL. It is about filtering out
> invalid syntaxes like:
> 
> -cpu ''
> 
> or
> 
> -cpu ,some-prop
> 

Greg, Thanks for your review on this trivial patch.

@cpu_option[0] is NULL when we have "-cpu ''". We run into assertion raised
by subsequent cpu_class_by_name(). However, @cpu_option[0] isn't NULL with
something like "-cpu ,xxx", but the CPU model specific class can't be found
at last.

So the validation mostly relies on cpu_class_by_name() if I'm correct. It's
fine to remove the check. However, it provides explicit error message, which
isn't bad though:

    error_report("-cpu option cannot be empty");

>> Signed-off-by: Gavin Shan <gshan@redhat.com>
>> ---
>>   exec.c | 5 -----
>>   1 file changed, 5 deletions(-)
>>
>> diff --git a/exec.c b/exec.c
>> index ffdb518535..3cff459e43 100644
>> --- a/exec.c
>> +++ b/exec.c
>> @@ -963,11 +963,6 @@ const char *parse_cpu_option(const char *cpu_option)
>>       const char *cpu_type;
>>   
>>       model_pieces = g_strsplit(cpu_option, ",", 2);
>> -    if (!model_pieces[0]) {
>> -        error_report("-cpu option cannot be empty");
>> -        exit(1);
>> -    }
>> -
>>       oc = cpu_class_by_name(CPU_RESOLVING_TYPE, model_pieces[0]);
>>       if (oc == NULL) {
>>           error_report("unable to find CPU model '%s'", model_pieces[0]);
> 

Regards,
Gavin

Re: [PATCH] exec: Remove the duplicated check in parse_cpu_option()

[Greg Kurz]

On Sat, 7 Dec 2019 23:56:55 +1100
Gavin Shan <gshan@redhat.com> wrote:

> On 12/7/19 3:58 AM, Greg Kurz wrote:
> > On Fri,  6 Dec 2019 17:33:37 +1100
> > Gavin Shan <gshan@redhat.com> wrote:
> > 
> >> The @cpu_option shouldn't be NULL, otherwise assertion from g_strsplit()
> >> should be raised as below message indicates. So it's meaningless to validate
> >> @model_pices[0] in parse_cpu_option() as it shouldn't be NULL either.
> >>
> >>     qemu-system-aarch64: GLib: g_strsplit: assertion 'string != NULL' failed
> >>
> >> This just removes the check and unused message.
> >>
> > 
> > Hrm... the check isn't about @cpu_option being NULL. It is about filtering out
> > invalid syntaxes like:
> > 
> > -cpu ''
> > 
> > or
> > 
> > -cpu ,some-prop
> > 
> 
> Greg, Thanks for your review on this trivial patch.
> 
> @cpu_option[0] is NULL when we have "-cpu ''". We run into assertion raised
> by subsequent cpu_class_by_name(). However, @cpu_option[0] isn't NULL with
> something like "-cpu ,xxx", but the CPU model specific class can't be found
> at last.
> 

You're right, the case with a leading ',' is caught by the other check.

> So the validation mostly relies on cpu_class_by_name() if I'm correct. It's
> fine to remove the check. However, it provides explicit error message, which
> isn't bad though:
> 
>     error_report("-cpu option cannot be empty");
> 

It's definitely not fine to remove an error message that clearly explains
to the user what he has done wrong in favor of QEMU aborting and printing
something cryptic like:

    cpu_class_by_name: Assertion `cpu_model && cc->class_by_name' failed.

Assertions are for bugs, not for bad command line usage.

> >> Signed-off-by: Gavin Shan <gshan@redhat.com>
> >> ---
> >>   exec.c | 5 -----
> >>   1 file changed, 5 deletions(-)
> >>
> >> diff --git a/exec.c b/exec.c
> >> index ffdb518535..3cff459e43 100644
> >> --- a/exec.c
> >> +++ b/exec.c
> >> @@ -963,11 +963,6 @@ const char *parse_cpu_option(const char *cpu_option)
> >>       const char *cpu_type;
> >>   
> >>       model_pieces = g_strsplit(cpu_option, ",", 2);
> >> -    if (!model_pieces[0]) {
> >> -        error_report("-cpu option cannot be empty");
> >> -        exit(1);
> >> -    }
> >> -
> >>       oc = cpu_class_by_name(CPU_RESOLVING_TYPE, model_pieces[0]);
> >>       if (oc == NULL) {
> >>           error_report("unable to find CPU model '%s'", model_pieces[0]);
> > 
> 
> Regards,
> Gavin
> 

Cheers,

--
Greg

Re: [PATCH] exec: Remove the duplicated check in parse_cpu_option()

[Gavin Shan]

On 12/8/19 3:51 AM, Greg Kurz wrote:
> On Sat, 7 Dec 2019 23:56:55 +1100
> Gavin Shan <gshan@redhat.com> wrote:
> 
>> On 12/7/19 3:58 AM, Greg Kurz wrote:
>>> On Fri,  6 Dec 2019 17:33:37 +1100
>>> Gavin Shan <gshan@redhat.com> wrote:
>>>
>>>> The @cpu_option shouldn't be NULL, otherwise assertion from g_strsplit()
>>>> should be raised as below message indicates. So it's meaningless to validate
>>>> @model_pices[0] in parse_cpu_option() as it shouldn't be NULL either.
>>>>
>>>>      qemu-system-aarch64: GLib: g_strsplit: assertion 'string != NULL' failed
>>>>
>>>> This just removes the check and unused message.
>>>>
>>>
>>> Hrm... the check isn't about @cpu_option being NULL. It is about filtering out
>>> invalid syntaxes like:
>>>
>>> -cpu ''
>>>
>>> or
>>>
>>> -cpu ,some-prop
>>>
>>
>> Greg, Thanks for your review on this trivial patch.
>>
>> @cpu_option[0] is NULL when we have "-cpu ''". We run into assertion raised
>> by subsequent cpu_class_by_name(). However, @cpu_option[0] isn't NULL with
>> something like "-cpu ,xxx", but the CPU model specific class can't be found
>> at last.
>>
> 
> You're right, the case with a leading ',' is caught by the other check.
> 
>> So the validation mostly relies on cpu_class_by_name() if I'm correct. It's
>> fine to remove the check. However, it provides explicit error message, which
>> isn't bad though:
>>
>>      error_report("-cpu option cannot be empty");
>>
> 
> It's definitely not fine to remove an error message that clearly explains
> to the user what he has done wrong in favor of QEMU aborting and printing
> something cryptic like:
> 
>      cpu_class_by_name: Assertion `cpu_model && cc->class_by_name' failed.
> 
> Assertions are for bugs, not for bad command line usage.
> 

Yes, Agree as explained previously. The explicit message is a bonus at least.
So please ignore this trivial patch and sorry for the noise.

>>>> Signed-off-by: Gavin Shan <gshan@redhat.com>
>>>> ---
>>>>    exec.c | 5 -----
>>>>    1 file changed, 5 deletions(-)
>>>>
>>>> diff --git a/exec.c b/exec.c
>>>> index ffdb518535..3cff459e43 100644
>>>> --- a/exec.c
>>>> +++ b/exec.c
>>>> @@ -963,11 +963,6 @@ const char *parse_cpu_option(const char *cpu_option)
>>>>        const char *cpu_type;
>>>>    
>>>>        model_pieces = g_strsplit(cpu_option, ",", 2);
>>>> -    if (!model_pieces[0]) {
>>>> -        error_report("-cpu option cannot be empty");
>>>> -        exit(1);
>>>> -    }
>>>> -
>>>>        oc = cpu_class_by_name(CPU_RESOLVING_TYPE, model_pieces[0]);
>>>>        if (oc == NULL) {
>>>>            error_report("unable to find CPU model '%s'", model_pieces[0]);
>>>

Regards,
Gavin