# Current sbuff failings
-# The major issues with the current sbuffs are
+## The major issues with the current sbuffs are
- Markers need to be released before the function they're created in returns. This means we often create temporary `fr_sbuff_t` on the stack which is inefficient.
- Using pointers often results in abuse, i.e. people storing pointers pointing into the underlying char buffer.
- Last matched token isn't stored anywhere, so the caller of a parsing function needs to check the current position of an sbuff against its tokens.
- Returning negative offsets to indicate error positions dosn't work when there are transform functions in the sbuff chain.
-# Fixes
-## Marker release
+## Fixes
+### Marker release
Current sbuffs require markers to be tracked because the marker's pointer needs to be updated is the sbuff is shifted,
} sbuff_marker_t;
----
-## Pointer abuse
+### Pointer abuse
In almost every instance a "legacy" C function that's operating on a standard C buffer takes a `buffer` and a `length`.
Standard copy functions could copy data from one "parsing" sbuff to a "printing" sbuff which wrapped a buffer on the stack.
Printing sbuffs could also be setup to do standard escaping for \0, making the string "c safe".
-## Overzealous pinning
+### Overzealous pinning
The key to fixing the pinning problem is to allow backtracking.
transform function is broken. i.e. if a transform function provides the `\` from `\t` in its output buffer, then the transform
function must be rewritten. The output buffer of a transform sbuff must contain only complete, transformed, atoms.
-## Shifting data out of sbuffs is expensive
+### Shifting data out of sbuffs is expensive
The reason why it's expensive is because the entire sbuff chain needs to have its pointers updated. If every sbuff maintains
positions using offsets, only the start and end offsets need to be changed. This gives a real advantage to using markers,
If `sbuff->offset < sbuff->start`, this event would trigger a backtrack.
-## Pasing terminal sequences and escape rules into functions is awful
+### Pasing terminal sequences and escape rules into functions is awful
Text transformation must be transparent to the parsing function, anything else will not function correctly, or will add
horendous complexity.
(we need to do that so it can find xlats). All unescaping and terminal sequence location is done transparently, and
`tmpl_afrom_substr` only sees the unescaped byte stream.
-## Negative offsets don't work when transform sbuffs are used
+### Negative offsets don't work when transform sbuffs are used
One key realisation is that returning negative offsets up the call stack will not work when the offsets don't map 1:1
between parent and child.
projects / thirdparty / freeradius-server.git / commitdiff
