Validate exports #357
Validate exports #357
Conversation
Codecov Report
@@ Coverage Diff @@
## master #357 +/- ##
========================================
Coverage 98.84% 98.84%
========================================
Files 38 38
Lines 11468 11571 +103
========================================
+ Hits 11335 11437 +102
- Misses 133 134 +1 |
gumb0
force-pushed
the
validate-exports
branch
2 times, most recently
from
38a45eb
to
05a3574
Compare
| throw validation_error{"invalid index of an exported function"}; | ||
| break; | ||
| case ExternalKind::Table: | ||
| if (export_.index != 0 || !module.has_table()) |
There was a problem hiding this comment.
Changed to using module class helpers
| module.imported_globals_mutability.size() + module.globalsec.size(); | ||
|
|
||
| // Validate exports. | ||
| std::unordered_set<std::string_view> export_names; |
There was a problem hiding this comment.
Is this much faster than vector or what is the reason?
There was a problem hiding this comment.
With set it is straight-forward to check if element exists.
lib/fizzy/parser.cpp
Outdated
| default: | ||
| assert(false); | ||
| } | ||
| if (const auto [_, inserted] = export_names.emplace(export_.name); !inserted) |
There was a problem hiding this comment.
I so much dislike this syntax of C++17 🤷
There was a problem hiding this comment.
if (!std::get<1>(export_names.emplace(export_.name)) isn't nicer.
But this entire thing is in a for loop, so why not just keep these as two statements (like the rest of codebase)?
There was a problem hiding this comment.
export_names.emplace(export_.name).second will work.
There was a problem hiding this comment.
changed it to @chfast's suggestion
| EXPECT_EQ(module.exportsec[3].name, "xyz"); | ||
| EXPECT_EQ(module.exportsec[3].kind, ExternalKind::Global); | ||
| EXPECT_EQ(module.exportsec[3].index, 0x45); | ||
| EXPECT_EQ(module.exportsec[3].index, 0); |
There was a problem hiding this comment.
Too bad the old test was invalid, but nicer to see what we're testing.
test/unittests/validation_test.cpp
Outdated
| @@ -231,3 +231,68 @@ TEST(validation, call_indirect_no_table) | |||
| const auto wasm = from_hex("0061736d0100000001050160017f00030201000a0901070020001101000b"); | |||
| EXPECT_THROW_MESSAGE(parse(wasm), validation_error, "call_indirect without defined table"); | |||
| } | |||
|
|
|||
| TEST(validation, export_invald_index) | |||
| (export "m" (memory 1)) | ||
| */ | ||
| const auto wasm_mem = from_hex("0061736d010000000503010001070501016d0201"); | ||
| EXPECT_THROW_MESSAGE(parse(wasm_mem), validation_error, "invalid index of an exported memory"); |
There was a problem hiding this comment.
Should we have different message whether memory/table section is present or not?
There was a problem hiding this comment.
I think this is fine - 0 is invalid index in some context.
|
Cases with exporting table/memory 0 when none is defined are missing from the spectests, added to the "to upstream" list. |
No description provided.