Check raw code length when instructions have same disassembly
but different encodings. Marks as OpMismatch to indicate
encoding difference.
Fixes#242Fixes#233
* Fix data flow analysis for multiple text sections
* Data flow analysis results were only keyed by the symbol (function)
address. That doen't work if there are multiple text sections, the
result from the first function in one section will stomp the result
from the first function in another because both have address zero.
* Remove the ambiguity by keying off of the section address as well.
* Formatting
* Satisfy wasm build
* Clippy
* Formatting again
* Thought that section was the section address not the section number.
---------
Co-authored-by: Luke Street <luke.street@encounterpc.com>
* WIP implementation
* * Move flow analysis to dedicated file
* Show string constants inline
* Handle calls to MWCC "sled" helpers which otherwise disrupt flow analysis
* Run cargo insta review
* Apply clippy feedback
* Update more tests.
* Remove std use from ppc flow analysis
* Try to make wasm build work again
* More test changes
* Probably last wasm fix
* Formatting
* Fix WASM
* One more clippy thing
* Fixed display of float constants in a LFS or LFD instruction in case where there is a branch to the subsequent instruction with a different register value.
* On lines with a reloc, only hide Symbol type data flow values rather than all data flow values.
* Formatting
* Use symbol name when comparing against an externed reloc
For partial matching files, often a symbol is externed even though it
should exist in the target object. We can still compare the symbol name,
instead of always returning a mismatch.
* Combine cases, and apply change reloc_eq in code.rs
When searching for a symbol by name, only look at
symbols that are defined within the object,
ignoring extern symbols (symbols without section).
Fixes#180Fixes#181
We have specific diff logic that relies on knowing
which object is the target object, and which is the
base. generate_mapping_symbols was designed in such
a way that it would reverse the target/base, leading
to a match percent shown that's different when it
gets applied.
Fixes#200
Reworks the local-branch handling logic to be more
unified: scan_instructions does all the work up front,
and process_instruction / display_instruction can
simply use the calculated branch destination instead
of performing their own is-relocation-target-
function-local checks.
(Hopefully) Fixes#192
* Reimplement colorized data relocation hover diffs
* Fix objdiff-wasm build
Data diffing doesn't seem to be fully implemented in objdiff-wasm yet, so just putting placeholders in so it compiles.
* Reloc hover: Add separators, override special color too
Before, this was comparing, for example, `.text-2`
with `.text-10` with standard string comparison,
yielding `.text-10` before `.text-2`.
Instead, this simply uses a stable sort by name,
which preserves the relative ordering of sections.
* PPC: Calculate pooled relocations
Reimplements #140
The relocations are now generated when the object is first read in `parse`, right after the real relocations are read.
`resolve_relocation` was changed to take `obj.symbols` instead of `obj` as an argument, because `obj` itself doesn't exist yet at the time the relocations are being read.
* Improve readability of PPC pool relocs code
* Fix regression causing extern pool relocs to be ignored
* Fix showing incorrect diff when diffing weak stripped symbol with an addend
This is a regression that was introduced by #158 diffing addends in addition to symbol names. But it's not really a bug in that PR, rather it seems like I simply never added the offset into the addend when creating a fake pool relocation for an extern symbol. So this commit fixes that root issue instead.
* Add PPC "Calculate pooled data references" option
* Fix objdiff-wasm compilation errors
* Update PPC test snapshots
* Show data literal values on instruction hover
Reimplements #108
* Show reloc diffs in func view when data's content differs
Reimplements #153
* Data diff view: Show relocs on hover and in context menu
This reimplements #154
Note that colorizing the text depending on the kind of diff has still not been reimplemented yet
* Fix up some comments
* Data view: Show data bytes with differing relocations as a diff
* Data view: Show differing relocations on hover
* Symbol list view: Adjust symbol/section match %s when relocations differ
* Improve data reloc diffing logic
* Don't make reloc diffs cause bytes to show as red or green
* Properly detect byte size of each relocation
* Data view: Add context menu for copying relocation target symbols
* Also show already-matching relocations on hover/right click
* Change font color for nonmatching relocs on hover
* Show reloc diff in func view when data content differs
* Add "Relax shifted data diffs" option
* Display fake pool relocations at end of line
* Diff reloc data by display string instead of raw bytes
This is to handle data symbols that contain multiple values in them at once, such as stringBase. If you compare the target symbol's bytes directly, then any part of the symbol having different bytes will cause *all* relocations to that symbol to show as a diff, even if the specific string being accessed is the same.
* Fix weak stripped symbols showing as a false diff
Fixed this by showing extern symbols correctly instead of skipping them.
* Add "Relax shifted data diffs" option to objdiff-cli
Includes both a command line argument and a keyboard shortcut (S).
* Remove addi string data hack and ... pool name hack
* Clippy fix
* PPC: Clear relocs from GPRs when overwritten
* PPC: Follow branches to improve pool detection accuracy
* PPC: Handle following bctr jump table control flow
* Clippy fixes
* PPC: Fix extern relocations not having their addend copied
* Add option to disable func data value diffing
* PPC: Handle lmw when clearing GPRs
* PPC: Handle moving reloc address with `add` inst
* Combine "relax reloc diffs" with other reloc diff options
* Add v3 config and migrate from v2
---------
Co-authored-by: Luke Street <luke@street.dev>
* Fix missing dependency feature for objdiff-gui
* Update .gitignore
* PPC: Display data values on hover for pools as well
* Tooltip data display: Format floats and doubles better
Floats and doubles will now always be displayed with a decimal point and one digit after it, even if they are whole numbers. Floats will also have the f suffix. This is so you can tell the data type just by glancing at the value.
* Move big functions to bottom ppc.rs
* Clear pool relocs in volatile registers on function call
This fixes some false positives.
* Revert ObjArch API changes, add fake target symbol hack
Because we no longer have access to the actual symbol name via sections, guess_data_type can no longer detect the String data type for pooled references.
* Add hack to detect strings via the addi opcode
* Move hack to resolve placeholder symbol into process_code_symbol
* Merge reloc and fake_pool_reloc fields of ObjIns
The Ghidra delinker plugin emits functions with type STT_OBJECT,
rather than STT_FUNC. The current logic was preventing these from
being compared based on their symbol type. Relax this condition
for now.
* Reduce dependencies for no features
* Add missing deps to every feature
* Add missing `dep:`s
* Gate even more deps behind features
Removes dependency on tsify-next / wasm-bindgen unless
compiling with the wasm feature by using `#[cfg_attr]`
* Fix wasm
---------
Co-authored-by: Luke Street <luke@street.dev>
This allows users to "map" (or "link") symbols with different names so that they can be compared without having to update either the target or base objects. Symbol mappings are persisted in objdiff.json, so generators will need to ensure that they're preserved when updating. (Example: d1334bb79e)
Resolves#117
