Try resolve #1175

[pavel-pimenov]

• New crash dump https://drdump.com/Problem.aspx?ProblemID=241665&Login=Guest
• Remove raw pointer and static std::string service_type;

[codecov-io]

Current coverage is 70.97% (diff: 77.77%)

Merging #1223 into master will decrease coverage by 0.13%

@@             master      #1223   diff @@
==========================================
  Files           390        390          
  Lines         58181      58182     +1   
  Methods        5728       5728          
  Messages          0          0          
  Branches       8779       8779          
==========================================
- Hits          41369      41294    -75   
- Misses        12443      12526    +83   
+ Partials       4369       4362     -7   

Powered by Codecov. Last update 85fe066...f516e47

[gubatron]

why not

if (d.service_namespace && !d.service_namespace.empty()) update_map(d, mapping_index);

and avoid a possible NPE?

The same comment applies everywhere else below, and since it happens so much perhaps I'd add a private utility function to not repeat yourself all over

bool valid_service_namespace(rootdevice &d) {
    return d.service_namespace && !d.service_namespace.empty();
}

and then have

if (valid_service_namespace(d)) update_map(d, mapping_index);

...

TORRENT_ASSERT(valid_service_namespace(d));

...

[pavel-pimenov]

service_namespace -> std::string
if(string()) - impossible

int main()
{
std::string name;
if(name) {
}
}

In function 'int main()': 8:10: error: could not convert 'name' from 'std::string {aka std::basic_string}' to 'bool'

[arvidn]

First of all, this fix should be made against RC_1_1, that's the version you're seeing this problem in, right?
Second; the existing code is clearly broken, I don't understand how it could get so screwed up. However, I'm tempted to fix it by restoring it instead of entirely switching to a std::string approach.
The original code (as far as I can tell from the intention of the existing code at least) meant for service_namespace to point to a string literal. There are only 3 choices of what this string can be, so that kind of makes sense. Alternatively, we could make it and index and have a lookup table for the actual namespace name.

[arvidn]

going back to RC_1_0 though, it doesn't look like this was ever right. going with std::string may be the better option

[arvidn]

this patch looks good to me. would you mind submitting it against RC_1_1?

[arvidn]

this has been merged into master from RC_1_1 now