[PATCH] D28569: Remove all variants of DWARFDie::getAttributeValueAs...() that had parameters that specified default values.

Wed Jan 11 15:06:35 PST 2017

I have submitted a new diff to review to see how we like the approach we discussed below:

https://reviews.llvm.org/D28581 <https://reviews.llvm.org/D28581>

Check out DWARFFormValue.h and look for the toString helper functions in there. One doesn’t take a default value and returns a Optional<const char *> and the other takes a default value and returns the default value if the value isn’t valid or isn’t a string.

Let me know what you think of this approach.

Greg Clayton

> On Jan 11, 2017, at 1:02 PM, David Blaikie <dblaikie at gmail.com> wrote:
> 
> 
> 
> On Wed, Jan 11, 2017 at 12:26 PM Adrian Prantl <aprantl at apple.com <mailto:aprantl at apple.com>> wrote:
> 
> > On Jan 11, 2017, at 12:07 PM, David Blaikie <dblaikie at gmail.com <mailto:dblaikie at gmail.com>> wrote:
> >
> >
> >
> > On Wed, Jan 11, 2017 at 12:00 PM Adrian Prantl <aprantl at apple.com <mailto:aprantl at apple.com>> wrote:
> >>
> >> > On Jan 11, 2017, at 11:33 AM, David Blaikie <dblaikie at gmail.com <mailto:dblaikie at gmail.com>> wrote:
> >> >
> >> >
> >> >
> >> > On Wed, Jan 11, 2017 at 11:08 AM Greg Clayton <clayborg at gmail.com <mailto:clayborg at gmail.com>> wrote:
> >> >>
> >> >>> On Jan 11, 2017, at 10:36 AM, David Blaikie <dblaikie at gmail.com <mailto:dblaikie at gmail.com>> wrote:
> >> >>>
> >> >>> Would it make sense to push the value retrieval down into DWARFFormValue to separate getting the value from a DIE from getting the value into some particular type - they seem pretty orthogonal & it's odd that DWARFDie handles the type handling side of that rather than an attribute/form abstraction.
> >> >>>
> >> >>> Perhaps it'd be a bit abusive, but we could even overload op[](dwarf::Attribute) on DWARFDie for easy map-like attribute lookup. Maybe 'find' would be a better name, though.
> >> >>>
> >> >>> I'm OK omitting the 'attribute' part of the name entirely if everyone's OK with overloading in this context & if the callsites tend to be pretty self documenting. (mostly if you're searching for a particular attribute you've probably got the attribute name in a literal at the callsite anyway, right?)
> >> >>>
> >> >>> So code like this, perhaps:
> >> >>>
> >> >>>   x = Die.find(DW_AT_name).asString();
> >> >>>
> >> >>> or perhaps:
> >> >>>
> >> >>>   x = asString(Die.find(DW_AT_name));
> >> >>>
> >> >>> (so that find/findAttr/[] whatever we call it can continue to return Optional<DWARFFormValue> which asString can pass along as NOne if it's passed None (or passed a non-string value), and otherwise decompose into a string, etc)
> >> >>>
> >> >>
> >> >> We basically have find already in:
> >> >>
> >> >> Optional<DWARFFormValue> DWARFDie::getAttributeValue(dwarf::Attribute);
> >> >>
> >> >> We can easily rename it.
> >> >>
> >> >> If find returns an optional then your first code example would need to be:
> >> >>
> >> >> x = Die.find(DW_AT_name).getValueOr(DWARFFormValue()).asString();
> >> >>
> >> >> The second form could be made to work but it doesn’t look as natural IMHO.
> >> >>
> >> >> Because of the first is awkward and the second seems less clear than just asking the DIE for the value I do think there is value in having the calls on DWARFDie. Let me know what you think.
> >> >
> >> > I'd still probably go towards the second - for orthogonality of features. Means a DWARFFormValue can be passed around and its value retrieved later, etc.
> >> >
> >> > But would be good to hear from Adrian too.
> >>
> >> No matter which version we end up using I would expect any heavy users of this to re-wrap it in something more specialized again so
> >>
> >>   x = asString(Die.find(DW_AT_name).getValueOr(DWARFFormValue()));
> >>
> >> becomes
> >>
> >>   x = getAttrAsOr<StringRef>(Die, DW_AT_name, "")
> >
> > Not sure I follow. I would figure:
> >
> >   x = asString(Die.find(DW_AT_name)).getValueOr("");
> >
> 
> Ok. That is different from how I understood the proposal, since asString here takes and returns an Optional. This also answers my other question, since we then can do getValueOr to unwrap the optional result of asString.
> 
> 
> > to be pretty readable/not be worth having wrappers, but I can see what you mean.
> 
> It's not awesome :-)
> The order of execution/reading is:
>   2            1                 3
>   asString(Die.find(DW_AT_name)).getValueOr("");
> 
> It's certainly not uncommon to mix member and non-member function calls in expressions... 
>  
> Sure it's readable, but it takes a second look to correctly parse it mentally.
> 
> Yep. Alternatively, if we had unique names for these functions (perhaps put them in a 'dwarf' namespace under llvm) we could potentially have:
> 
> Optional<const char*> toString(const DWARFFormValue&)
> 
> Die.find(DW_AT_name).apply(toString).getValueOr("");
> 
> perhaps (adding an Optional<T>::apply(Functor) function - but it'd have to collapse Optionals on the return value). It'd be nicer because toString wouldn't have to support an Optional parameter, but would add some complexity to Optional (& really 'apply' should probably be a free function since it's not in std::optional - which would defeat some of the point of the readability improvements).
>  
> 
> > I'm not sure if there would be a nice way to keep the orthogonality (so that asString/asAddress/etc didn't all need to support/reimplement default handling) while making it much more terse. It seems like there are 3 different things happening there.
> 
> Would it make sense to provide both variants?
> I think it would be fine to compose the terse variant out of the generic orthogonal building blocks.
> 
> We certainly could. It wasn't clear to me if it was worth adding the wrappers for each of these. Whether shaving off a few characters at the call sites helped that much.
>  
> 
> >
> > But, yes, if it comes up often enough to be annoyingly verbose - sacrificing the orthogonality (by creating/paying in the form of duplication) for convenience by writing asFoo functions that handle the defaulting - asString(Die.find(DW_AT_name), "") - same as we had before, except now the conversion is orthogonal to the lookup so adding "bool recurse"/"find_recursively" doesn't explode the API surface area along another axis.
> >
> >> to reclaim some readability again. From that point of view I don't think the decision of where asString lives is very important. I am wondering how to pass any default value other than an empty value into getValueOr(DWARFFormValue Default)? Is it easy to construct an on-the-fly DWARFFormValue for e.g., a string?
> >
> > I don't think we'd want people to construct DWARFFormValues to represent values that weren't present in the DWARF as a main/common means of handling situations like this. That seems to be the wrong layer - I think at the point you want an appropriately typed value, then you collapse the optionality and values pace into that type if you want to, until then it remains Optional.
> >
> 
> Yes, I think I was misguided by the
>   asString(Die.find(DW_AT_name).getValueOr(DWARFFormValue()))
> example.
> 
> -- adrian
> >>
> >> >
> >> >
> >> > Another option that Adrian might have thoughts on, would be to remove the Optional wrapper around the DWARFFormValue from getAttributeValue. It's not something I'd be totally happy with - Optional's nice and explicit about the fact that there might not be a value, but if DWARFFormValue already has a "not present" state in it, we could arguably use that state to indicate that.
> >>
> >> I think the answer to this depends on my previous question.
> >> -- adrian
> >

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170111/6c8e0752/attachment.html>