On Saturday, 15 December 2018 at 21:09:12 UTC, Sebastiaan Koppe wrote:
> On Saturday, 15 December 2018 at 15:37:19 UTC, Basile B. wrote:
>> I think this is what Walter calls "AST poisoning" (never understood how it worked before today). And the whole parser is like this.
>>
>> This poisoning kills the interest of using a fuzzer. 99% of the crashes will be in hdrgen.
>
> As is common with fuzzing, you'll need to ensure the program crashes.

Yes this is done by piping dmd with the random code (i dont use dmd as a library for now). If the process returns something different of 0 (ok) and 1 (normal compiler error) than the random code is saved in a file:

        ...
        ProcessPipes pp = pipeProcess([Options.dc, "-"]);
        pp.stdin.writeln(src);
        pp.stdin.close;
        if (!pp.pid.wait.among(0, 1)) fileName.write(src);
        ...

Actually it would be less convenient to do that with the front end as a library, since SEGFAULTs are supposed to kill the program...

On 2018-12-15 12:29, Basile B. wrote:
> While the D front end is not yet used to make tools
I've used it to make a tool, DLP [1].

[1] http://github.com/jacob-carlborg/dlp

-- 
/Jacob Carlborg

On 2018-12-15 16:37, Basile B. wrote:

> This poisoning kills the interest of using a fuzzer. 99% of the crashes will be in hdrgen.

Does that matter as long as the bug is found?

-- 
/Jacob Carlborg

On Sunday, 16 December 2018 at 14:24:54 UTC, Jacob Carlborg wrote:
> On 2018-12-15 16:37, Basile B. wrote:
>
>> This poisoning kills the interest of using a fuzzer. 99% of the crashes will be in hdrgen.
>
> Does that matter as long as the bug is found?

Well it's hard to tell if it's begin.
Generally in a compiler which is focused on speed, you accept crashing an really bogus input if it makes the parser run faster, because there is no chance of accepting corrupted code, which is what you need to be worried about.

On Monday, 17 December 2018 at 10:12:44 UTC, Stefan Koch wrote:
> On Sunday, 16 December 2018 at 14:24:54 UTC, Jacob Carlborg wrote:
>> On 2018-12-15 16:37, Basile B. wrote:
>>
>>> This poisoning kills the interest of using a fuzzer. 99% of the crashes will be in hdrgen.
>>
>> Does that matter as long as the bug is found?
>
> Well it's hard to tell if it's begin.

meant to say benign.