DSDP/DD/Face-to-face Toronto 22-Feb-2006
- 1 Agenda & Attendee List
- 2 Presentations
- 3 Minutes - Wednesday DD Meeting
- 4 Minutes - Thursday TM/DD joint session
Agenda & Attendee List
Minutes - Wednesday DD Meeting
Eclipse 3.2 Debug Platform - Demos / feedback session on prototyping
- Pawel Piech - Wind River
- Implemented directly against platform
- Pawel P - PDA Prototype
- The current state of the flexible hierachy aligns well with WR's debugger implementation.
- Main issues at this point:
- Need retargetable actions
- Need public interface to get at standard images for label providers
- Column support in views is incomplete at this point
- Alan Boxall
- Implemented directly against platform
- Moving from 3.0 to 3.1 - the biggest challenge was to get the multi-threaded UI to talk to their synchronous debug engine.
- They queue up all asynchronous requests. They use a lot of caching.
- They are using the compatibility mode right now.
- Taking advantage of 3.2 EDM in the future: eventually, IBM's debug engines will drive the hieararchy.
- Number of jobs is somewhat alarming. (See performance discussions below.)
- Plan to do some prototyping against flexible hierarchy after 3.2 is released.
- Mikhail Khodjaiants, QNX
- Mikhail K - CDT Flexible Hiearchy Update
- CDT has requests to provide customized versions of variables and registers view. Probably will happen after CDT 3.2.
- Using compatibility mode right now for flexibile hierarchy (for 3.1). Still need more investigation for how to expose customization.
- We need flexible hierarchy exposed at CDI layer.
- We use disassmebly view. Need a disassembly memory renderer.
- Summary, they need flexibility at top and bottom.
- They would like to see CDT define a more embedded-centric user experience without major changes to CDI.
- Some view customizations.
- Multi-core flexibility is important.
- Similar comments to TI.
- Builds against Eclipse platform, but haven't had a chance to look at 3.2 yet.
- Builds against Eclipse platform.
- Migrated an old VB debugger. Looked at CDT and did some prototyping, but decided it would be too much work.
- They are lacking some features using the platform directly, but also believe they have less problems. One big issue was using GDB with their architecture.
- First product: Oct 05 and based on 3.1.
- Biggest issue is trying to use the memory view with their architecture.
- Haven't prototyped against 3.2 EDM yet, but they can benefit from simplified hiearchy. Update policies are also critically important because of very slow target connections. Will focus on 3.2 after April.
- Continually re-evaluate CDT. Could potentially use parts of CDT.
- Have a released product on 3.0 with a customized CDT.
- Working on a 3.1-based product with un-modified CDT. Trying to use GDB.
- Also like view customization.
- TI: Demo of view customizations in E 3.1
- Variables and registers view are in table tree format (using tree control)
EDM 3.2 Progress Update – Darin W (IBM)
- Stuff in M5 (public API freeze)
- Virtual tree - turned out to be more complicated than they thought.
- Virtual table
- Incremental load
- Stuff not in M5
- Scopes / drag and drop
- Moving forward, our API's are provisional, so we can still make changes.
- March 30 - M6 feature freeze - only 3 weeks of development left
- Columns and removing the last remaining synchronous interfaces
- What's next after the 3.2 release?
- How much feedback they get will determine how quickly these provisional API's can be public
- Would need feedback in first 3 months: July - Sept
- The customized view content may have to go through another cycle.
- 125374 - multi-column support in variables view: patch posted. Darin says this is close to the actual design. Ted would like more granuality in toggling specific columns on and off. Ted will add this to bug entry.
- Plugable cell editors probably won't make the release. Ted to look at provided a patch to possibly help get this in.
- API versioning - adding new interfaces on top of old ones vs. deprecating. When are interfaces collapsed together
- Intention is that the adapters would live for a while in order to enable the backward-compatible debug model.
- Using the editor with multiple debugging backends
- Double-click gutter action - how is this resolved when the same editor is shared between two debug engines.
- Seems to be a bigger problem that spans multiple projects: CDT, DD, PTP
- IBM will come to EclipseCon that shows performance across multiple versions
- Concerns about the large numbers of working threads spawned
- Will have some suggestions for performance improvements
- Action items: each company should run its own performance numbers
- Progress Update – Samantha (IBM)
- Samantha C - Update Policy
- Slide 4: Element, Model Proxy Factory, Model Proxy, and Model provided by client. "Update Policy" is actually "View Updater"
- Implementation is in the model and is model-specific.
- Still work to do before we have a generic implementation.
- Update policy will have to be part of viewer, and model proxy will have to be part of the model.
- Samantha C - Update Policy
- Need to form a workgroup to address the issues in Samantha's presentation. Need to collect use cases from folks in this group.
- Need to share a common look-and-feel, even if the implementation is in the model.
- What's in 3.2
- You can create model Proxies that tell how and when debug events are handled
- Not there: a generic update policy
- Demo by Ted Williams (WR) of new renderer for memory view.
- DD group would like this contributed
- Other features
- Programmatic foreground and background coloring
- Customizable context menu
- Symbol interleave with address bar
- Optional confirm before write
- Undo write
- Column headings with sub-address - user can turn on and off
Contributions and Participation Discussion
- See Felix's notes
- Proposal: Technology sub-groups
- Purpose is to create faster progress on the many technical areas the group wants to purpose.
|Debug view||Paul Gingrich (TI)||TBD|
|Memory||Samantha Chan (IBM)||Ted Williams (WR), Freescale, AMI|
|Expressions & Variables||Ken Ryall (Nokia)||Ted Williams (WR), Pete Nichols (IBM)|
|Registers||Kirk Beitz (Freescale)||Ted Williams (WR), Pete Nichols (IBM), ATI|
|Breakpoints||Ewa Matejska (PalmSource)||Pawel Piech (WR), Mikhail Khodjaiants (QNX), TI, Nokia|
|Console - Debug/Serial/Network||Aaron Spear (Mentor)||Freescale, AMI, PalmSource|
|Disassembly||Chris Recoskie (TI) - temporary lead||Wind River|
|Views - multi-context, pin/clone, update policy||Darin Wright (IBM)||Samantha Chan (IBM), Ted Williams (WR)|
|Debug Model||Pawel Piech (Wind River)||Mikhail Khodjaiants (QNX), Freescale, Nokia|
|Launching||Pete Nichols (IBM)||TI|
|Editor||Ewa Matejska (PalmSource)||Pete Nichols(IBM), Nokia|
|Source Lookup||Pawel Piech (Wind River)||Pete Nichols(IBM)|
- Leader drives discussions and prototyping on their technology. Leader initiates conference calls, start a wiki page (see link above), collects requirements, and investigates/delegates prototyping. Leader is responsible for making sure the discussion is progressing. Leaders have commit rights to DD.
- Team members help provide requirements and help protoype.
- All communications should happen on dsdp-dd-dev mailing list for visibility.
- By next meeting: Lead should have requirements at minimum, but should also have protoype if at all possible, since having something to look at will generate better feedback.
- Current developer commitments (used to build table above)
- 1 half-time engineer
- Plan to update memory rendings, registers view, and variables view over the course of this year. Targeting Eclipse 3.1.2. right now management. Eclipse 3.2 product will release late summer / early fall.
- 2 half-time+ engineers
- Picking up 3.2 in fall release.
- Still working on first eclipse based product, so not yet ready to commit to providing fixes for any issues. After first release, they will have more breathing room for possible contributions.
- Also releasing first product, but could probably get some resources approved if community can delegate some work, especially in CDT. Looking for direction and needs from Mikhail.
- ATI (Mentor)
- Still evangelizing internally. SPIRIT participation is definitely possible. Potentially some tool contributions for creating, parsing, and checking SPIRIT files.
- Really want to contribute, but have no resources yet. Could possibily contribute a debug console.
- Would like to contribute work on the views. Currently working on second Eclipse product. Looking at adding extensible breakpoint behavior and possibly some work on variable detail formatting for c++. After that they'll look at feedback from customers to help set priorities.
- 1 engineer
- Memory update policies and possibly contribute inbound launch code. Need to think about other areas.
- How can we get better participation to help Darin out on the debugger interfaces and views?
- Where do we want to go next? Volunteers for implementation?
- New breakpoint features?
- More memory rendering?
- Sample debugger implementation from Wind River?
- Debug console
- Committer List
- This list is based on the sub-project leads who volunteer to build use-cases and coordinate prototyping for platform improvements. The commiters will have edit access to the Wiki and DSDP/DD CVS repository.
- Paul Gingrich (TI)
- Samantha Chan (IBM)
- Ken Ryall (Nokia)
- Kirk Beitz (Freescale)
- Ewa Matejska (PalmSource)
- Aaron Spear (Mentor)
- Chris Recoskie (TI)
- Darin Wright (IBM)
- Pawel Piech (Wind River)
- Pete Nichols (IBM)
- Ted Williams (WR)
- Eclipse 3.2 launching framework feedback
- Darin described the launch changes in 3.2, which are covered in "Integrating Custom Debuggers into the Eclipse Platform" - his tutorial at EclipseCon 2006.
- Next Meeting - Late May / Early June. Topics
- Technology sub-group updates
- If face-to-face, do a possible coding camp to implement some prototypes and enhancements.
Minutes - Thursday TM/DD joint session
Update on DSDP, Plans for EclipseCon - Doug G
- See DSDP Website. Discussed new sub-projects: MTJ and NAB. Project pages will be posted in 1 to 2 weeks.
- All companies will be represented at EclipseCon. Aaron S, Mikhail K, Kirk B, Tom H won't be there, though.
- AI: Doug to setup lunch tables - due by Mar 1. Will coordinate with Doug S to not duplicate CDT areas.
- BOF - schedule at the conference.
- Hobson - Introduction
- SPIRIT for EDA - this is more for information
- SPIRIT for Debuggers
- Debug working group is probably needed to drive the SPIRIT definition.
- Ideal if Eclipse could join the consortium. ARM and Mentor can be interfaces, too.
- Hobson could work on setting up the debug working group. Hobson will also check on what they can make available for parsers. Hobson will check on licensing of SPIRIT XML files.
- Next steps
- Each member company to look into joining SPIRIT
- DSDP-DD and DSDP-TM to nominate representative for a debug working group
- DSDP-DD and/or DSDP-TM to build tools for generating/reading/validating SPIRIT files. AI for Aaron to talk to company about potential contributions. No Java Parser available yet for SPIRIT - we could contribute this in DD.
- Doug - WR's data files standards
- Two feature requests from SPIRIT
- Board initialization specification
- Help file indexing so registers can point to Silicon Vendor's pdf/html manuals
- Flow: SoC describes silicon -> Board vendor describes hardware and initialization -> user customizes initialization.
- Two feature requests from SPIRIT
- Aaron Spear - general discussion on SPIRIT requirements
- Target Descriptions in SPIRIT
- Feedback on additional register spec requirements
- Unique ID
- Initial grouping, but ability for a user to change groups
- Textual descriptions - problematic for I18N - how should SPIRIT handle?
- Mapping between debug format and registers - does this belong in SPIRIT or elsewhere?
- Disassembly information should be a part of this - eventually we'd like table-driven disassembly - separate discussion
- Endianness, default display size: byte, word, long, double, etc.
- Help index
- Long description for tool tip
- Feedback on additional memory spec requirements
- Flash - should be a separate discussion - CFI standard (?)
- Virtual - should be a separate topic - ATI limited the XML to what's physically on the chip. - need to describe MMU structure
- Need to think about run-time memory map changes that are tied to the OS. Debugger will need to look at the MMU structure and also need to know about kernel data structures to handle the effective-to-physical address translation.
- Shared memory between cores