This is a partial revert of the type id change (bd2653fb) to get
the case of virtual bases work again, as it falls back to native
field accesses everywhere we have native types.
This is likely to be a performance regression with CDB/Python,
but since this combination is experimental and not used by
default in 4.2 getting GDB/LLDB right is more important.
Medium term we might want to have a more fine-grained approach
to use of type ids (e.g. use native field accesses on a case-by
-case base only for type with virtual inheritance)
Change-Id: I239111e6ce5f3365b750068bfc4dafb12be1d53c
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
Use the field's type size if the reported field size is 0.
Change-Id: I7db49507d0ea7358b6db27dfa206f2b079bbf876
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
There seems to be race in the LLDB setup on Linux. Work around
by not doing serious work twice to be able to use the test at least.
Change-Id: I387a57d3c66e6c4209bd424af9be16f17acb98fa
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
Move all the currentFoo initialization to DumperBase
Change-Id: I0a59715fe1550f466e78cbf678622dd8dfe1783c
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
Reviewed-by: David Schulz <david.schulz@qt.io>
Reviewed-by: hjk <hjk@qt.io>
Instead, push backend differences into a more generic reportResult()
function.
Change-Id: I6bc34580e915b830e592680022b3499ae5eb1e57
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
Lowers the barrier for other debugger backends.
Change-Id: I09e0ad09548b6b4220175245cc0d845ac5aa29d0
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
Reviewed-by: hjk <hjk@qt.io>
It's not the one of the lldb process anymore, so set it explicitly.
Change-Id: Ibcc2b63631c4096f860996ad6ccfc12415d8aefd
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
GDB and LLDB share some ideas but are not identical. Map them
both to what we need in the end. This also reduces the size
of the Type class interface as a step towards more type info
caching.
Change-Id: I43e60c4e5736ac15a5c776832ffb8e6939f45e45
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
Easier maintenance if code is typesafe. Whatever that means in Python.
Change-Id: I450e5d42bcfdb30d607dd878353d037ba18fe6d9
Reviewed-by: hjk <hjk@qt.io>
lldb stops the process after attaching and the event loop is not yet started.
This makes the debugger wait for events while process is in stopped state
Task-number: QTCREATORBUG-15705
Change-Id: Iae6fe94fc483d963b377582c4cbbb443be5e6cba
Reviewed-by: hjk <hjk@qt.io>
This introduces a dumper types abstraction layer with classes
Dumper.{Value,Type,Field} wrapping either gdb.{Value,Type,Field}
or lldb.{SBValue,SBType,SBField} and uses it to move
to more direct memory accesses in the dumper implementation.
This way we can use duck typing for artificial intermediate
objects eliminating the need for {gdb.Value,lldb.SBValue}.cast()
in some case which are flaky in general and typically not
available in release builds.
As consequence QRegion and QVariant dumper work without debug
info now.
Change-Id: Iea2411175ef67f2bf651ee7eaade9879ed5ceba1
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
Previously if program execution was paused, and a c++ command was
executed via executeDebuggerCommand method, which in turn printed
to stdout, Qt Creator would assume that program execution has
been resumed. But in reality the execution was still paused.
Make sure not to report the program execution as being resumed, in
case if something is printed to stdout as a result of a debugger
command.
Change-Id: I8752be00b1bf5bd4767debc2eb26b9a433f251bb
Reviewed-by: hjk <hjk@qt.io>
Make QMeta* dumper work without debug info.
Show QObject parent and children list in this case.
QObject properties are not visible, and neither actual type
of QObject children.
Change-Id: I05417a7d78084234403a964f683db6867b6686f0
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
At least LLDB reports sometimes out of scope variables as such.
Report them as "out of scope" instead of letting the common
struct dumper fail and report its generic "not available".
Change-Id: Ic9d0d196894ef811fab8e6a586bf55d80cd02e45
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
map() directly returned a list in python 2.
Fixes a startup issue on windows.
Change-Id: Ib4b67ca98a74135ff108546f8d089686f7df9493
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
Commands needs to be given in Python, i.e. something like
import time
print time.asctime()
for i in range(1, 10):
print(i)
works.
Task-number: QTCREATORBUG-15585
Change-Id: I7724617e4bbe85a717ae78b21014e2b55c4089c8
Reviewed-by: Alexandru Croitor <alexandru.croitor@theqtcompany.com>
Reviewed-by: hjk <hjk@theqtcompany.com>
They may contain quotes and possibly other data breaking the transport
protocol.
Change-Id: I53795bd1de54385e8d9c50088368fdc768db217b
Task-number: QTCREATORBUG-15635
Reviewed-by: Alexandru Croitor <alexandru.croitor@theqtcompany.com>
Reviewed-by: hjk <hjk@theqtcompany.com>
Backends only have to specify whether an object members are sortable
in principle (e.g. all structs), and some numeric 'sortgroup' value
for member items (higher values are always sorted on top).
Change-Id: I10ce94580374fed48a35f058a575a1408d6801af
Reviewed-by: David Schulz <david.schulz@theqtcompany.com>
Reviewed-by: Christian Stenger <christian.stenger@theqtcompany.com>
LLDB 3.8 on Linux claims non-existent files to be the source of
disassembly.
Change-Id: I45a47cf5e7faad7ccd6dbb9879c8328fc2496b95
Reviewed-by: Christian Stenger <christian.stenger@theqtcompany.com>
* Update license of files loaded in debugger
* Make sure all files have a license header
Change-Id: I612885652bcae0a4b7a88ae5184ef661aee25006
Reviewed-by: hjk <hjk@theqtcompany.com>
At least on Linux/LLDB 3.8 trying to set watchpoints at fixed address
produces breakpoints with .IsValid() == False that choke on accessing
certain functions like .IsOneShot() used e.g. in breakpoint data
reporting.
Now report the .IsValid() state and skip the reporting of the "bad"
fields.
Change-Id: I742ed171f37763f3dbbd896173c527817edf3a6e
Reviewed-by: Alexandru Croitor <alexandru.croitor@theqtcompany.com>
Reviewed-by: Christian Stenger <christian.stenger@theqtcompany.com>
LLDB on Linux did not like a plain toInteger() in some cases.
Change-Id: I4fb640df52218a3df04f3ce5092eb5fbc8b5acd5
Reviewed-by: Christian Stenger <christian.stenger@theqtcompany.com>
... into a string specifying a kind of encoding (utf8, latin1),
an optional bytesize (2-byte integers, 4-byte integers) and a
flag whether the displayed value should get "..." around it.
Scales better than adding an enum value for each new combination.
Change-Id: Iffcb1e2f148f12da96e165559a976bd34026c649
Reviewed-by: David Schulz <david.schulz@theqtcompany.com>
... and switch it off by default. The feature has indeterministic
performance impact and lost quite a bit of its utility since GDB
learned to extract dynamic object types most of the times.
Change-Id: I22cccb03ba67f9ff6ad757bbc06eb372d84bbffe
Reviewed-by: Leena Miettinen <riitta-leena.miettinen@theqtcompany.com>
Reviewed-by: hjk <hjk@theqtcompany.com>
Apple's LLDB has changed behavior between 330.x and 340.x.
Newer versions seem to actively delete the DYLD_IMAGE_SUFFIX,
DYLD_LIBRARY_PATH and DYLD_FRAMEWORK_PATH environment variables
from the scripting environment.
Change-Id: I0b7bbdce0975e969b1dae5770fd9a99328f241be
Reviewed-by: Eike Ziller <eike.ziller@theqtcompany.com>
... in the presence of empty base classes. This now actually makes
the Inheritance dumper test pass.
Change-Id: I92ada5da46accb6e7eaff58ce5dc329484ef0909
Reviewed-by: Christian Stenger <christian.stenger@theqtcompany.com>
... as well as the 'autoderef' flag. A lot of dumper tests were
failing even though the dumpers themselves were ok.
Change-Id: I198784d95eb0004e81bfab995121e048b1722c66
Reviewed-by: Christian Stenger <christian.stenger@theqtcompany.com>
Empty base classes are counted towards GetNumberOfDirectBaseClasses()
but do not get reported in GetChildCount(). Fix the difference by
manually creating empty base class entries in the report.
Change-Id: I2650d9f5234302c5bf2fa0e62a886d3a39fc11db
Reviewed-by: Christian Stenger <christian.stenger@theqtcompany.com>
CamelCase might be more Qt-ish, but the backends tend to use
lowercase only.
Change-Id: I04b9b7305b54226f27b70151115050c4816f911f
Reviewed-by: Christian Stenger <christian.stenger@theqtcompany.com>
