Re: error messages

[Nicholas Blumhardt]

Hi Brad,

Thanks for the feedback- glad Sprache is working for you.

The feature you're looking for is already implemented in the source repository but hasn't been built into a new release. Checking out the code and building it yourself is probably the fastest way to go, another official release is still a little while away.

Cheers,
Nick

On 28/04/2011, at 2:58 AM, BradW <bra...@gmail.com> wrote:

> Hello, Sprache is a life saver for us! I’ve re-written our DSL parser
> with it, and it’s a great improvement. Our parser is now 250 lines of
> readable rules, replacing 500 lines of spaghetti code in our hand-
> rolled parsing class.
>
> Now I am wanting to show better error messages to the user, especially
> showing the line number rather than the char position number. Is there
> any documentation or example of producing more information in error
> messages?
>
> Thanks,
> Brad

[Brad Williams]

Hi Nicholas,

Thanks. I ended up parsing the character count out of error message
and then compute the line number from that, not ideal but fairly
simple anyway. Now that I have your source will look for where you
report the error line number.

Next issue on error messages: I was able to get the error messages to
appear as desired on the smallest granular parser by using XMany().
However, this is forcing me to change my language so that two elements
that can appear in lists ("Questionnaire" and "QuestionGroup") will
NOT both start with "Q". So changing the second to just "Group"
solved that.

Of course this is not ideal to have to change the language definition.
I am wondering what you think of making a String parser (maybe call
it XString) to be more aggressive than just finding a "partial match"
on the first character, and instead require that it either match the
*entire string* or return a failure. I was trying to come up with
alternate XString parser to act this way, but I don't know if it will
work -- I don't understand enough of how this code works. Do you
think this is not going to work out, like there is no way to create an
XString parser to work this way? (Warning: I haven't even run this
code, but I think you see what I'm trying to accomplish...to do NO
i.Advance() unless we know that the entire string I'm looking for is
going to match...)

public static Parser<IEnumerable<char>> XString(string s)
{
return i =>
{
if (!i.AtEnd)
{
var remainingString = i.Source.Substring(i.Position);

if (remainingString.Length >= s.Length)
{
if (remainingString.Substring(0, s.Length) == s)
{
for (int c = s.Length; c > 0; c--)
i.Advance();

return new Success<IEnumerable<char>>(s, i);
}
}

return new Failure<IEnumerable<char>>(i,
() => string.Format("unexpected '{0}'", i.Current),
() => new[] { "expected " + s });
}

return new Failure<IEnumerable<char>>(i,
() => "Unexpected end of input reached",
() => new[] { "expected " + s });
};

[BradW]

Okay, I got around to actually trying my code above ... it works with
one change. In the for loop the i must be replaced with the return
from Advance() like this:

i = i.Advance();

Nicholas, can you please make the Success and Failure classes public?
This will allow me to define my own parsing matcher such as XString
outside of the sprache.dll.

Thank you for this great project!

Brad Williams

[Nicholas Blumhardt]

Hi Brad,

Glad to hear it is coming together!

I'm keen to see this supported; it looks to me like it would be nicer to provide some more 'combinatorial' way of doing it - e.g. something like Parse.Fail() in:

public static Parser<IEnumerable<char>> XString(string s)

{

   return Parse.String(s)

      .Or(Parse.Fail<IEnumerable<char>>())

      .Named(s);

}

Parse.Fail() would just return the 'unexpected ...' at the input position where the XString() parser starts, e.g. along the lines of:

public static Parser<T> Fail<T>()

{

   return i => new Failure<T>(i,

                                          () => string.Format("unexpected '{0}'", i.Current),

                                          () => Enumerable.Empty<string>());

}

Sound like it would work? I think I've seen something similar done before.

My time for updating Sprache is a bit limited at the moment, if you are able to raise this in the issue tracker I'll try to make some progress as soon as possible. Alternatively if you are interested in implementing something like Parse.Fail and some associated tests I'd be happy to give you access to do that.

Regards,
Nick