Re: 11.0_BETA i386 dosbox segfault

[Ramiro Aceves <ea1abz%gmail.com@localhost>]

El 16/11/25 a las 13:29, Mike Pumford escribió:
> On 15/11/2025 09:28, Ramiro Aceves wrote:
> >
>
> > Http_connect_socket connect ERROR: No route to host
> > We will try another IP address.
> > ctype(3) isspace: invalid input: -62
> > [1]   Abort trap (core dumped) dillo
> > packardbell_netbsd$
> This one at least is easy to explain. NetBSD 11 picks up on applications 
> calling the C isxxx functions like isspace() here with arguments that 
> cause undefined behaviour. So the bug here is in dillo which appears to 
> think that passing -62 to that function is valid (which it is not).
>
>  From memory there is an environment variable you can set to make the 
> system tolerate these errors but the real fix here is to get dillo fixed 
> if possible as on pretty much all systems passing -62 to that function 
> could cause a crash or other horribly random behaviour anyway.
>
> I suspect if you do man ctype it will tell you what the workround 
> environment variable is.
>
> Mike


Yes, David told as how to "fix" it with LIBC_ALLOWCTYPEABUSE="TRUE" 
variable. Thanks for pointing to the man page:

ENVIRONMENT
     In NetBSD 11, the ctype functions will always crash with a signal on
     certain invalid inputs as a diagnostic aid for applications; see 
CAVEATS.
     Setting the environment variable LIBC_ALLOWCTYPEABUSE before 
starting a
     program will restore the old behavior of returning nonsense 
answers for
     these inputs, or sometimes but not always crashing, depending on 
factors
     such as address space layout randomization.


Thanks so much!

>