Commit Graph
195 Commits
Author SHA1 Message Date
Nikita SmithandRyan Fleury 65288208af dbg_engine: extend trap-net range across contiguous same-source-line ranges 2026-06-02 12:10:27 -07:00
Nikita SmithandRyan Fleury d7e5c5a830 dbg_engine: bugfix memory fetch in step_over, step_over_inst, and step_into
and incorrect boolean logic for trap-net
2026-06-02 12:10:27 -07:00
Nikita SmithandRyan Fleury 61724e88ad event shadowing in dbg engine ctrl causes debugger to return incorrect
last event
2026-06-02 12:10:09 -07:00
Nikita SmithandRyan Fleury 4648b27a06 increase wait time for memory slice in line stepping 2026-06-02 12:10:05 -07:00
Nikita SmithandRyan Fleury b23323c03d add switch to override default logs folder 2026-06-02 12:10:04 -07:00
Ryan Fleury 95f4ade10d lift wait-for-process-memory-cache-update out of d_process_write, since that path runs on control thread now too as the general 'process handle write' abstraction - just move into the rd_eval_space_write path, since that wait was only for UI benefit anyways 2026-05-19 14:35:29 -07:00
Ryan Fleury 1a48155116 checkpoint on dump file parsing / cache construction; temporarily turn off crash dump command, since still wip 2026-05-19 09:56:01 -07:00
Ryan Fleury fe4de192e4 linux -> prefix user raddbg data with ., so we get ~/.raddbg normally - on windows keep normal style 2026-05-18 20:18:53 -07:00
Ryan Fleury 40885a3f59 fix unmapped address visualization when reading is impossible + stop it from terminating rest of useful visualization 2026-05-18 20:01:15 -07:00
Ryan Fleury 80d68b7f95 dbg_engine: minidump memory64 parsing; rename switch -> better thing for what it is 2026-05-18 19:40:47 -07:00
Ryan Fleury 4840b46b84 dbg_engine: shift thread register reads / process memory reads -> abstraction layer, rather than directly calling into demon; cache crash dump info, like thread info & memory ranges; implement thread reg reads & memory reads for dumps by using this acceleration structure when we have a dump handle, instead of regular demon reads. 2026-05-18 14:51:17 -07:00
Ryan Fleury 6d6cd7033d minidumps: initial gather / events for process/modules/threads 2026-05-16 20:53:46 -07:00
Ryan Fleury fddf353ea0 minidump format types; open crash dump command, stub out opening path in control thread; expand d_handles to support machine id, entity id, *and* 'controller kind', which can be either a dump file or a demon - this properly separates the true process control abstraction (demon) from opening a regular dump file on any machine, which will let us trivially support any kind of dump file on any platform; also fix bad cfg tree state leak 2026-05-16 18:26:23 -07:00
Ryan Fleury 58caf0d3d5 upgrade eval space reads with staleness/bad/change info, to expand process memory cache reading abilities to all evaluation spaces; unmapped address visualization (both memory view & single line eval) 2026-05-13 14:48:00 -07:00
Ryan Fleury a1778421c9 eval extended opcode for setting base offset - used for correct cross-module evaluations, fixes go-to-definition when thread is not in module of resolved identifier's symbol 2026-05-13 11:23:48 -07:00
Ryan Fleury 264abe57aa allow missing debug info in case of 'step out' - we still want to support traditional behavior especially in e.g. disassembly and so on 2026-05-13 08:40:14 -07:00
Ryan Fleury d45090a8cc adjust kerning to be rounded, rather than floor'd; adjust step-out desc 2026-05-11 19:53:54 -07:00
Ryan Fleury d1536953b0 p2r: eliminate spurious duplicate address vmap entries; upgrade step-out with block stepping functionality 2026-05-11 14:21:43 -07:00
Ryan Fleury de35a86505 fix artifact cache replaced-value deletion - this CAN NOT run if the new value == the old value 2026-05-11 13:29:29 -07:00
Ryan Fleury 52da228efe artifact_cache: fix leak of old-generation artifacts, when those artifacts are replaced by newer generation computations. artifacts can opaquely hold things like arena pointers, and so when we replace them in the cache, we need to respect the lookup api's provision of create/destroy hooks for the cache, such that the artifacts can be released correctly (using the destroy hook) before being replaced. this is separate from regular eviction. 2026-05-11 11:49:15 -07:00
Ryan Fleury d59d6a21f0 d2r2: do not duplicate line records, were being generated by columnar changes, but these are not currently supported by the debugger 2026-05-09 12:18:05 -07:00
Ryan Fleury aea4a4ff9d eliminate os_core, move into base; eliminate regs, move into arch abstraction layer, define x64 regs in x64 layer; move win32 backend for os core (now base) functionality to top-level win32 layer 2026-05-08 15:21:21 -07:00
Ryan Fleury 2eb0b8b233 ctrl -> d 2026-04-22 14:53:37 -07:00
Ryan Fleury 0e3c821e28 collapse ctrl/d bps 2026-04-22 14:31:08 -07:00
Ryan Fleury 0d8a6a4d3a ctrl -> d type pass 2026-04-22 13:37:00 -07:00
Ryan Fleury a90fbc844b merge ctrl_init / d_init 2026-04-22 12:56:15 -07:00
Ryan Fleury 2388db3998 fold ctrl layer into dbg engine - no benefit to disallowing dependence on debug engine types inside of ctrl, and it has been causing unnecessary barriers for eventual eval magic. 2026-04-22 12:44:18 -07:00
Ryan Fleury be270a8f8d fix step over/into line operations when source line maps to multiple (potentially) discontiguous code ranges, or at least line info records; this fixes weird stepping with Clang-generated line info for for loops, function prologues in Odin, amongst other similar issues 2026-04-08 14:42:30 -07:00
Nikita Smith 20d3fa4ddb normalize paths with inconsistent slashes 2026-02-20 22:08:26 -08:00
Nikita Smith f9561cb618 add support for CFA in the eval system
Canonical Frame Address is a DWARF-only concept encoded in the unwind info.
Debugger must store this data for each call frame and provide it to
the eval system based on selected stack frame.

To abstract over format details, we introduce a "Frame Unwind Context".
For DWARF, this context maintains the intermediate unwind state and exposes
the CFA, whereas for PE, it remains empty.
2025-11-26 15:42:40 -08:00
Nikita Smith 0565fdef0a WIP add support for GNU LibC TLS model
Unlike on Windows, TLS index isn't part of the module format. glibc stores
it in the opaque link_map struct. There are many different heuristics
a debugger can use to locate it but most of them are unreliable and
prone to errors, or don't work when debugger attaches to a process.
We adopt the newer approach of looking for special symbols
in glibc that provide TLS index offset. These symbols were added in
2021, so programs linked with older glibc won't have TLS support.
2025-11-21 16:41:35 -08:00
Ryan Fleury c8c25c0f98 processless debug info loading/unloading; separate modules from dbg infos in eval; keep dbg infos around after debugging via config, correllate to new modules, evict when necessary, when generating new versions 2025-10-22 16:39:29 -07:00
Ryan Fleury 92517ef2d7 re-enable single-step on empty trap net, if no line info is present 2025-10-14 10:00:31 -07:00
Ryan Fleury ec69ebff35 fix tick-deferred command mechanism in debug engine 2025-10-07 15:50:52 -07:00
Ryan Fleury 783988c0ac expand ctrl process memory cache access to allow waiting for fresh results, rather than any results; use when doing a ctrl process memory write, to prefer viewing less stale data on exact frame. also expand trap net generation with success bit, use to retry stepping commands across frames if process memory reads are hindered. abort after N retries. 2025-10-07 15:44:35 -07:00
Ryan Fleury a58c23754a di2 -> di; delete old dbg info layer, delete async layer 2025-10-01 17:30:08 -07:00
Ryan Fleury 4964a94cd3 transition to dbg_info2 2025-09-30 12:52:32 -07:00
Ryan Fleury ad9c67ae43 eliminate old caches from ctrl layer 2025-09-26 13:44:19 -07:00
Ryan Fleury ca7bfab7ea bucket artifact cache requests by wideness/priority; do high/wide, high/thin, low/wide, low/thin 2025-09-24 17:08:08 -07:00
Ryan Fleury 5381307e90 hash_store -> content 2025-09-18 14:42:25 -07:00
Ryan Fleury 937c6c8bce eliminate radcon, eliminate unused string helpers 2025-09-16 10:17:02 -07:00
Ryan Fleury 056604fb94 sketch out tline -> vline text wrapping cache types; first pass of async call stack tree builder 2025-08-11 13:46:06 -07:00
Ryan Fleury 190b3384f9 preserve selected thread info on continues 2025-07-16 15:52:41 -07:00
Ryan Fleury ce9f7f691a always-refresh dev option 2025-06-20 10:55:45 -07:00
Ryan Fleury 401f117653 adjust step trap net build in case of memory read failure - ensure we get correct memory read, otherwise we can generate bad trap nets during heavy load. allow high vs. low priority in async rdi loading; use low-priority in pre-emptive case 2025-06-18 08:36:41 -07:00
Ryan Fleury c282e7d4be command outputs; ipc command output reception & writing to stdout 2025-06-05 16:08:53 -07:00
Ryan Fleury 94b969f203 copyright year 2025-06-02 14:13:57 -07:00
Ryan Fleury 58df68b9be adjust good-machine-code check to factor in memory read failure mask 2025-05-22 17:15:51 -07:00
Ryan Fleury 4aeadc12b7 adjust memory reading fail conditions in trap net generation; fix early-aborting of retry loop in process memory cache lookup 2025-05-22 12:15:31 -07:00
Ryan Fleury 8bed56adde step-into-line trap nets: omit stopping at non-final call destinations with no line info 2025-05-20 16:15:03 -07:00