FunctionDepsFinder: Fix closure computation not analyzing functions which have a prototype declared after its body #28
Conversation
Previously, if a struct was defined as:
struct S {
} __aligned(64);
whereas __aligned(x) is a macro, clang-extract would only copy
__aligned, thus misscompiling the code.
Signed-off-by: Giuliano Belinassi <[email protected]>
…mbol Previously, FunctionDepsFinder::Find_Functions_Required sometimes only analyzed the last declaration of a function, which breaks in the case where the function has a new prototype defined later of its body. The reason for this is that DeclContext::lookup fails to find all symbols of a given name. If this is a result of a bug in LLVM this is yet to be determined. Fixes SUSE#20. Signed-off-by: Giuliano Belinassi <[email protected]>
Signed-off-by: Giuliano Belinassi <[email protected]>
giulianobelinassi
force-pushed
the
closure_fix
branch
from
dd53351 to
ba20722
Compare
|
Thanks, it fixes the issue! Some interesting notes here: Like stated on #5, some macros can be redeclared: Not issue for now, it can be addressed by the livepatch developer. Another interesting result of this patchset is that it now adds multiple declarations of a function. For example: /** clang-extract: from drivers/net/wireless/intel/iwlwifi/iwl-dbg-tlv.h:56:1 */
void klpp__iwl_dbg_tlv_time_point(struct iwl_fw_runtime *fwrt,
enum iwl_fw_ini_time_point tp_id, union iwl_dbg_tlv_tp_data *tp_data, bool sync); was found on line 758, then /** clang-extract: from drivers/net/wireless/intel/iwlwifi/iwl-dbg-tlv.c:1347:1 */
void klpp__iwl_dbg_tlv_time_point(struct iwl_fw_runtime *fwrt,
enum iwl_fw_ini_time_point tp_id, union iwl_dbg_tlv_tp_data *tp_data, bool sync) was found on line 2027, but now implementing the function, and then we have: extern void klpp__iwl_dbg_tlv_time_point(struct iwl_fw_runtime *, enum iwl_fw_ini_time_point, union iwl_dbg_tlv_tp_data *, bool);On line 2122, which seems like what klp-ccp currently does. Either way, now it generates code that compile, which is the most important part, but I would like to point to these details to have it documented somewhere. Merge it! |
|
Actually, on klp-ccp, instead of the |
This may be a bug in the PrettyPrint::Print_Macro_Undef, so it may be worth investigating.
|
clang has a mechanism (DeclContext::lookup) which is SUPPOSED TO return the list of all Decls that matches the lookup name. However, this method doesn't work as intented. In kernel (see github issue #20) it results in the lookup_result set not having the Decl that has the body of the function, which is the most important!
Therefore, we fallback to the older but slower method: sweep through the AST top level looking for the Decls we want to extract.
If this issue in the symbol name hash comes from a bug in LLVM, this still needs investigation.