Better html links #793
Better html links #793
Conversation
|
Ah, it was an assert crashing, |
|
There you go. Fixed and working with |
|
Wouldn't the links be even nicer if they look like this: |
|
And even nicer if they sorted alphabetically for each term rather than the current mess. The changes so far were to make hrefs unique and the index look is just a side-effect of it not using anything more elaborate than the href to display to the user, since the I think http://dl.dropboxusercontent.com/u/145894/t/ndocs/rstgen.html#setIndexTerm could accept an additional default parameter with the wanted look for the index hyperlink. Do you know if it is safe for the index files to grow a third column to store it? Is there any other program reading index files? |
|
I don't think the index file is meant for anything besides humans at the moment, but @Araq may know better. |
The .idx file format was extended to contain two columns more, one with the text of the href, one with a title which browsers render as popups.
|
Bumped with better index. |
|
Replaced by #850. |
## Summary Move the routines and types that make up the DSL to the `mirconstr` module and use the DSL in both the `injectdestructors` and `mirbridge` module. Apart from making the MIR code generation in both modules less complex, manual, and error-prone, this is also significant progress towards the MIR code coming out of the `injectdestructors` pass having proper type information, which is a requirement for having MIR passes that run after `injectdestructors`. ## Details * move the `EValue` and `ChainEnd` types to `mirconstr` * move the atomic DSL operands to `mirconstr`. The `notOp`, `modify`, `outOp`, and `tupleAccess` operands stay in `mirgen` * add wrapper templates for DSL operands that `mirgen` still calls with a `TCtx` instance instead of a `MirNodeSeq` As part of moving them, some operand routines that had a 'gen' prefix are changed to not have it, making the routine names more consistent. In addition, some cleanup in `mirconstr` is performed, and the DSL extended a bit: * `|=>`, `=>|`, and `previous` are removed. They were unused and didn't fit in with the overall design * the `emit`, `symbol`, and `opParam` operands are added, all of which are only useful outside of `mirgen` Another addition is the `predicate` meta operand, which makes it possible to write chains where an operand is conditionally excluded: ```nim chain(...): a() => predicate(cond) => b() => c() ``` Here, `b()` is only evaluated if `cond` evaluates to 'true'. Using two chains plus an in-between 'if' statement was previously necessary to achieve the same. Finally, documentation about the DSL is added, and the `injectdestructors` and `mirbridge` modules are changed to use the DSL for generating MIR code. The logic from the `genInjectedSink` procedure was previously duplicated into `genSinkFromTemporary`, but this is now fixed.
The following changes work fine in
doccommands except for the lack of types when a parameter is found with a default parameter but without a specified type. When I try to run indoc2mode I don't seem to make sense of thesymstructure and the proc to convert thetypto a string sigsevs without a call stack, so I don't know where what is failing.