Case insensitive Identifiers

[Milian Wolff]

David Nolden, 15.02.2010:
> Am Montag 15 Februar 2010 16:27:18 schrieb Niko Sams:
> > On Mon, Feb 15, 2010 at 16:00, David Nolden <zwabel at googlemail.com> wrote:
> > > Am Montag 15 Februar 2010 14:50:34 schrieb Niko Sams:
> > >> On Mon, Feb 15, 2010 at 13:13, David Nolden <zwabel at googlemail.com>
> 
> wrote:
> > >> > Am Montag 15 Februar 2010 12:49:50 schrieb Niko Sams:
> > >> >> Hi all (David),
> > >> >> 
> > >> >> In Php class identifiers and function identifiers as case
> > >> >> insensitive. Currently there are two problems
> > >> >> with that:
> > >> >> - QuickOpen displays the tolower version
> > >> >> - I want to change code completion to use CodeModel only (for
> > >> >> performance) and i'ts tolower then too
> > >> >> 
> > >> >> two ideas how to solve this:
> > >> >> 1. add a CaseInsensitive flag to findDeclarations
> > >> > 
> > >> > That's not possible, because the global PersistentSymbolTable does
> > >> > not support case-insensitive look-up, and it would probably be a
> > >> > major amount of work to make that work.
> > >> > 
> > >> >> 2. when showing QuickOpen/CodeCompletion items load shown
> > >> >> declarations
> > >> > 
> > >> > At least for quick-open, this shouldn't be a problem, as it happens
> > >> > anyway! Quickopen loads all the declarations that are currently
> > >> > visible completely, so it can disambiguate multiple overloads of the
> > >> > same function, and such stuff. You'd just have to modify it so it
> > >> > also extract the non-tolower name when loading the declarations. I
> > >> > think the place to do this is
> > >> > DUChainItemData::text.
> > >> 
> > >> Ah, ok this patch would fix the issue:
> > >> --- a/kdevplatform/plugins/quickopen/duchainitemquickopen.cpp
> > >> +++ b/kdevplatform/plugins/quickopen/duchainitemquickopen.cpp
> > >> @@ -46,7 +46,7 @@ QString DUChainItemData::text() const {
> > >> 
> > >>        if(def->declaration())
> > >>        
> > >>            decl = def->declaration();
> > >> 
> > >> -  QString text = decl->qualifiedIdentifier().toString();
> > >> +  QString text = decl->toString();
> > > 
> > > I think that won't work, because we need the whole qualified
> > > identifier, and toString() only returns the local identifier (we need
> > > 'Namespace::Name' and not just 'Name')
> > 
> > Ah, ok, Php doesn't have namespaces :D
> > Alternative solutions?
> 
> But you also have "ClassName::functionName", no? There you'll need it as
> well. So you probably have to traverse the scope somehow.
> 
> You could also implement a "virtual QString
> Declaration::prettyQualifiedIdentifier() const { return
> qualifiedIdentifier().toString() }" and override it in your php stuff to
> return the non-tolower version.

While this should theoretically work, it would also require that we store a 
pretty qualified identifier. Right now we only store a pretty identifier, and 
afaik the qualified identifier is created automatically for us somewhere in the 
abstract *builder, not sure though. I'd prefer to prevent unloading, see 
below:

> > >>    if(!decl->abstractType()) {
> > >> 
> > >> Is is Ok?
> > >> 
> > >> >> This is not easily possible with CodeCompeltion as kate uses the
> > >> >> NameColumn and DisplayRole for sorting
> > >> >> 
> > >> >> 
> > >> >> Any ideas how to solve this isue?
> > >> > 
> > >> > Hmm in code-completion it's more problematic, that's true. So you
> > >> > don't want to use declarations any more at all? At least in C++, the
> > >> > declarations are used anyway so they can be filtered and such, and
> > >> > in that case it wouldn't be a problem to also extract the
> > >> > non-tolower name at that point.
> > >> 
> > >> I do want to use them, but they should lazy load - as currently when
> > >> using a large Php framework thousands of declarations each one in it's
> > >> own top context gets loaded - and that takes here 15sec+.
> > >> So my idea is to load the declaration only when it's visible, and use
> > >> the data from CodeModel for the others. But the problem is sorting
> > >> that currently used the same role & column as for display is used.
> > > 
> > > But it's only that slow when completing for the first time, no?
> > 
> > Yes. But that doesn't help, it takes ages - it needs to be improved.
> 
> But at some point you will need those declarations anyway. You could also
> simply preload all those declarations in the background once you open the
> first file.

How can we prevent the DUChain cleanup thread kicking in and unloading the 
DUContexts again? Is it enough to keep a Pointer on them? Is there something 
like a DUContext::Ptr or similar?

And a general question: Having a Context doesn't mean that all containing 
items are stored as well, or does it? I mean when we only need global stuff 
available all the time, would this than also mean that every local var is in 
memory?

-- 
Milian Wolff
mail at milianw.de
http://milianw.de
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part.
URL: <http://mail.kde.org/pipermail/kdevelop-devel/attachments/20100215/c856a568/attachment.sig>