Consider granting the Context class virtual functions

[voetsjoeba]

I'm currently experimenting with using the Grantlee_Templates library as a replacement for the (very) basic templating language that the Shibboleth SP project provides to render HTML error report pages:

https://wiki.shibboleth.net/confluence/display/SHIB2/NativeSPErrors

It's working well so far, but the issue that I'm running into is that the templating API of the Shibboleth SP project uses dynamic variable lookup, i.e. there is a TemplateParameters class with a getParameter(const char* varName) method that gets invoked at runtime to look up the desired variable from a set of different possible sources in a well-defined order.

In my replacement POC I have access to such a TemplateParameters instance, but I can't "plug it in" to the Grantlee templating system because Context::lookup is declared as non-virtual, so I have no chance of modifying how variables are looked up at runtime. If instead it were declared virtual, I could create a subclass of Context that dispatches runtime lookup of variables to the TemplateParameters instance (though only after first failing the original Context::lookup --the ability for local variables to be pushed onto the stack at runtime by various tags like for and with must stay respected).

So my question is this: would you consider virtualizing Context::lookup and related others to allow end users to customize runtime variable lookup?

[steveire]

That could be considered for a Qt 6 based Grantlee, but not before then because of binary compatibility.

However, why can't you pre-generate the context data instead of using a request-like API as you describe? Is it mostly for porting convenience?

[voetsjoeba]

Because I have no reliable way of precomputing all variables. The getParameter function internally looks up variables from a number of different sources to which I have no direct access. I can look at the source code and find out and hack in manual access to those sources and replicate the final list of variables it would produce, but that requires hacking and will break as soon the implementation of getParameter changes.

[dantti]

This is also why IMHO having the same version as Qt jumping from 0.x to 5.x was a bad idea, can't a different named method be added as virtual?