| 1 |
# Error recovery (2022-05-07) |
| 2 |
|
| 3 |
Problem: we have two fairly generic cases of recovery bounded within a range: |
| 4 |
- sequences: `int x; this is absolute garbage; x++;` |
| 5 |
- brackets: `void foo() { this is garbage too }` |
| 6 |
|
| 7 |
So far, we've thought of these as different, and had precise ideas about |
| 8 |
brackets ("lazy parsing") and vague ones about sequences. |
| 9 |
Con we unify them instead? |
| 10 |
|
| 11 |
In both cases we want to recognize the bounds of the garbage item based on |
| 12 |
basic token-level features of the surrounding code, and avoid any interference |
| 13 |
with the surrounding code. |
| 14 |
|
| 15 |
## Brackets |
| 16 |
|
| 17 |
Consider a rule like `compound-stmt := { stmt-seq }`. |
| 18 |
|
| 19 |
The desired recovery is: |
| 20 |
- if we fail at `{ . stmt-seq }` |
| 21 |
- ... and we can find for the matching `}` |
| 22 |
- then consume up to that token as an opaque broken `stmt-seq` |
| 23 |
- ... and advance state to `{ stmt-seq . }` |
| 24 |
|
| 25 |
We can annotate the rule to describe this: `{ stmt-seq [recovery] }`. |
| 26 |
We can generalize as `{ stmt-seq [recovery=rbrace] }`, allowing for different |
| 27 |
**strategies** to find the resume point. |
| 28 |
|
| 29 |
(It's OK to assume we're skipping over one RHS nonterminal, we can always |
| 30 |
introduce a nonterminal for the bracket contents if necessary). |
| 31 |
|
| 32 |
## Sequences |
| 33 |
|
| 34 |
Can we apply the same technique to sequences? |
| 35 |
Simplest case: delimited right-recursive sequence. |
| 36 |
|
| 37 |
``` |
| 38 |
param-list := param |
| 39 |
param-list := param , param-list |
| 40 |
``` |
| 41 |
|
| 42 |
We need recovery in **both** rules. |
| 43 |
`param` in the first rule matches the *last* param in a list, |
| 44 |
in the second rule it matches all others. We want to recover in any position. |
| 45 |
|
| 46 |
If we want to be able to recovery `int x int y` as two parameters, then we |
| 47 |
should extract a `param-and-comma` rule that recovery could operate on. |
| 48 |
|
| 49 |
### Last vs not-last elements |
| 50 |
|
| 51 |
Sequences will usually have two rules with recovery, we discussed: |
| 52 |
- how to pick the correct one to recover with |
| 53 |
- in a left-recursive rule they correspond to last & not-last elements |
| 54 |
- the recovery strategy knows whether we're recoverying last or not-last |
| 55 |
- we could have the strategy return (pos, symbol parsed), and artificially |
| 56 |
require distinct symbols (e.g. `stmt` vs `last_stmt`). |
| 57 |
- we can rewrite left-recursion in the grammar as right-recursion |
| 58 |
|
| 59 |
However, on reflection I think we can simply allow recovery according to both |
| 60 |
rules. The "wrong" recovery will produce a parse head that dies. |
| 61 |
|
| 62 |
## How recovery fits into GLR |
| 63 |
|
| 64 |
Recovery should kick in at the point where we would otherwise abandon all |
| 65 |
variants of an high-level parse. |
| 66 |
|
| 67 |
e.g. Suppose we're parsing `static_cast<foo bar baz>(1)` and are up to `bar`. |
| 68 |
Our GSS looks something like: |
| 69 |
|
| 70 |
``` |
| 71 |
"the static cast's type starts at foo" |
| 72 |
---> {expr := static_cast < . type > ( expr )} |
| 73 |
| "foo... is a class name" |
| 74 |
+---- {type := identifier .} |
| 75 |
| "foo... is a template ID" |
| 76 |
+---- {type := identifier . < template-arg-list >} |
| 77 |
``` |
| 78 |
|
| 79 |
Since `foo bar baz` isn't a valid class name or template ID, both active heads |
| 80 |
will soon die, as will the parent GSS Node - the latter should trigger recovery. |
| 81 |
|
| 82 |
- we need a refcount in GSS nodes so we can recognize never-reduced node death |
| 83 |
- when a node dies, we look up its recovery actions (symbol, strategy). |
| 84 |
These are the union of the recovery actions for each item. |
| 85 |
They can be stored in the action table. |
| 86 |
Here: `actions[State, death] = Recovery(type, matching-angle-bracket)` |
| 87 |
- we try each strategy: feeding in the start position = token of the dying node |
| 88 |
(`foo`) getting out the end position (`>`). |
| 89 |
- We form an opaque forest node with the correct symbol (`type`) spanning |
| 90 |
[start, end) |
| 91 |
- We create a GSS node to represent the state after recovery. |
| 92 |
The new state is found in the Goto table in the usual way. |
| 93 |
|
| 94 |
``` |
| 95 |
"the static cast's type starts at foo" |
| 96 |
---> {expr := static_cast < . type > ( expr )} |
| 97 |
| "`foo bar baz` is an unparseable type" |
| 98 |
+---- {expr := static_cast < type . > (expr)} |
| 99 |
``` |
| 100 |
|
| 101 |
## Which recovery heads to activate |
| 102 |
|
| 103 |
We probably shouldn't *always* create active recovery heads when a recoverable |
| 104 |
node dies (and thus keep it alive). |
| 105 |
By design GLR creates multiple speculative parse heads and lets incorrect heads |
| 106 |
disappear. |
| 107 |
|
| 108 |
Concretely, the expression `(int *)(x)` is a valid cast, we probably shouldn't |
| 109 |
also parse it as a call whose callee is a broken expr. |
| 110 |
|
| 111 |
The simplest solution is to only create recovery heads if there are no normal |
| 112 |
heads remaining, i.e. if parsing is completely stuck. This is vulnerable if the |
| 113 |
"wrong" parse makes slightly more progress than the "right" parse which has |
| 114 |
better error recovery. |
| 115 |
|
| 116 |
A sophisticated variant might record recovery opportunities and pick the one |
| 117 |
with the rightmost *endpoint* when the last parse head dies. |
| 118 |
|
| 119 |
We should consider whether including every recovery in the parse forest might |
| 120 |
work after all - this would let disambiguation choose "broken" but likely parses |
| 121 |
over "valid" but unlikely ones. |
| 122 |
|
| 123 |
|