 Rajat Dua
|
8c8bf79f24
Removing -force:fieldhoist and -off:fieldhoist from rlexe.xml files
|
7 年之前 |
 Irina Yatsenko
|
8a219b6d94
Add CollectGarbage to the global object only in chakracore builds and only if requested
|
8 年之前 |
 Doug Ilijev
|
b6629367a1
Remove unused 'bug' and 'bugfix' test tags.
|
8 年之前 |
 Akrosh Gandhi
|
bf6ae644b6
[1.5>1.6] [MERGE #3341 @akroshg] 17-07 ChakraCore servicing release
|
8 年之前 |
 Paul Leathers
|
2e514d7265
[CVE-2017-8609] On object pointer copy prop, check for mismatch between new pointer symbol and existing object type spec info. If there is a mismatch, then stale ObjTypeSpecFldInfo may lead to jitted code that accesses a bad address, so disable optimization of this load/store.
|
8 年之前 |
|
Doug Ilijev
|
2709dd5272
Remove defunct exclude_ship from test tags.
|
8 年之前 |
|
Doug Ilijev
|
69761809dd
Add bug references.
|
8 年之前 |
|
Doug Ilijev
|
a63e04d508
Fix up rlexe.xml files: Remove BOM; remove spaces between tags which cause tags to be missed under rl.exe.
|
8 年之前 |
|
Doug Ilijev
|
d982c8472e
Remove baselines that only contained 'pass' since that's handled in the test runner.
|
9 年之前 |
 Derek Morris
|
48ff0d4599
Add test for ldlen regopnd jitoptimized fix.
|
9 年之前 |
 Michael Holman
|
412b9983ce
add missing serialization layouts; disable forceserialized from some UTs
|
9 年之前 |
 Jianchun Xu
|
50bb88eebf
xplat: fix rlexe.xml filename case
|
9 年之前 |
|
Paul Leathers
|
8567a69a3c
Types that are used in equivalent objtypespec guards must be locked to protect against, e.g., adding a property to a dictionary type handler, which will only change the type if it's locked.
|
10 年之前 |
|
Paul Leathers
|
d62f9505da
Even in cases where there's another type check upstream (so the operand is marked type-available) we may do a monomorphic type check on the spot (e.g., fixed field check). So instead of checking type-available to see whether we need to push a monomorphic guard type up past a given instruction, look to see whether the instruction has a type check bailout.
|
10 年之前 |
|
Paul Leathers
|
bd254ebae3
Old issue in forward pass objtypespec analysis exposed by recent backward pass fixes. If we have a monomorphic check upstream and reach a point where a monorphic check expects a different type, we can't replace the checked type in the value info. We have to indicate a type mismatch.
|
10 年之前 |
|
Paul Leathers
|
c36d2eaedb
Since we don't want to perform monomorphic type checks at points where the type is already available, avoid clearing the monomorphic type check bucket in such a case.
|
10 年之前 |
|
Paul Leathers
|
8271639b91
Installation of the monomorphic guard type check should only happen at the original type check that guards the property access requiring monomorphism, as indicated by non-availability of the checked type.
|
10 年之前 |
|
Paul Leathers
|
67a4f6b27d
My previous fix to monomorphic type checks had a subtle mistake. To handle the case where a downstream store adds a property, I grabbed the store's cached initial type to use as the guard type at the monomorphic check. But in some cases, the type we checked upstream is the final type, not the initial one. One-line fix for that.
|
10 年之前 |
|
Paul Leathers
|
9b49b23fa6
With equivalent obj type spec on by default, certain operations still require monomorphic type checks. We've handled this in the dead store pass by locating any equivalent type check that protects an operation that requires a specific type and changing it to use a monomorphic check. But because of the vagaries of inline cache sharing, the operation that does the check may not appear to be monomorphic. Deal with this by having the dead store pass indicate the required type on the checking instruction.
|
10 年之前 |
 Louis Lafreniere
|
1ed98f6dff
Objptr copy-prop bug
|
10 年之前 |
 Curtis Man
|
801556d3e5
Add UTF8 BOM to test xml files and make line endings consistent for test js and baseline files
|
10 年之前 |
 ChakraBot
|
5d8406741f
Initial commit
|
10 年之前 |
