Release Notes — CCStudio™ IDE v20.5.0

Need ability to sort names alphabetically. This will enhance usability for long register lists.

Need ability to automatically save build output to a log. This was supported with CCS Eclipse but not in CCS Theia.

Support export to file like we do in memory browser for registers view

We should enable command history in the new scripting console that persists across CCS launches.

Provide ability to use CCS IDE variables inside CCS Scripting

If I have a project that is set to use TI Arm Clang and I change it to use the older TI Arm Compiler we seem to lose all settings. And by all I don't just mean compiler options. Product dependencies are lost. i.e. if the project had a dependency setup for an SDK and SysConfig that is gone. SysConfig options are cleared. User compiler and linker options are gone. While it is understood that some options may not migrate, even ones that are the same between the tool chains (i.e. include) are lost.

The clang compiler option -iquote is an alternative to -I. It adds a entry to the list of directories searched for files included with the syntax ... {noformat} #include "filename.h" {noformat} Use -iquote on directories that contain a file with the same name as a standard header file, but is not a replacement for that file. Add a way for a CCS project to specify a directory with -iquote.

The registers view does not have support for a float format type. Both "Real" and "Scientific" are for decimal. It would make more sense to have those for float like CCS Eclipse.

There should be an UI to guide the user to easily configure the available/supported AI tools for CCS. * Provide a list of tested AI extensions that works for CCS the user can install * Have an UI or instruction to guide the user to configure the installed extension * Provide an UI to guide the user to configure Theia AI with the required information using the standard AI model and customer model

Have a debug MCP tool call to start a projectless debug session.

This is to simplify user's workflow when switching sysConfig MCP versions. Advanced users can choose to ignore the prompt and just restart the MCP connection.

Create a MCP server that can be used with any LM, such as Claude, Cline, etc.

Create a SysConfig MCP server (https://modelcontextprotocol.io/docs/getting-started/intro) so that code-assistance agents (Theia AI, windsurf, copilot, etc) will be able to modify SysConfig settings through the tool. Without this, these tools often try to modify the .syscfg script directly which frequently leads to mistakes that are difficult for the user to debug. By going through SysConfig, there will be validation and user feedback as to what is being done.

Hardware breakpoints are only available on the first device of multiple daisy chained devices. There are no issues with software breakpoints. Other debug actions like connect, load, run, read/write registers/memory, etc also work fine. Operations like source stepping code in flash (asm stepping is fine) can be impacted since that action could require a breakpoint. There are no issues at all with the first device. Note that bypassing the first device so that only the second device is debuggable does not help. It appears only the device physically first on the scan-chain is unaffected by this issue. It is reproducible with a variety of CCS versions, starting with 9.x to 20.x, and debug probes (XDS100v2/v3, XDS110, XDS200). This issue does not appear with CCS 8.3.1 and earlier.

Disassembly view doesn't decode a VDIV.F64 instruction in a TMS570LC43x (Cortex-R5 big endian) program. It is simply displayed as: ".word 0xee810b00". The Cortex-R5 is in ARM mode and changing the Disassembly Style from "Mixed ARM/Thumb" to "ARM Only" didn't make a difference. Looking at the assembler listing saved by the compiler shows the expected VDIV.F64 instruction with the encoding of EE810B00.

CCS20 variables and watch views show incorrect values for PRU core. When you hover over variable name in the Variable or Watch views, it shows address 0x000000@Program Memory, the displayed value is also incorrect (since it's read from that address). CCS Eclipse also shows incorrect value from Program Memory but the address is correct

User settings don't get preserved after CCS update. Workspace level settings are not affected. Previously installed extensions may also be impacted.

Watchpoint "Data Mask" and "Data Value" fields for C2000 do not work as expected. A simple example where only bit 0 is toggled and the rest of the bits are cleared, the value of the variable changes from 0 to 1. The expectation of setting a mask of 0x1 and a data value of 0x0 is that the watchpoint would trigger only when the variable value is 0 with bit being cleared. However the watchpoint never triggers despite this being the case. The data value is set to 0x1 (with the mask remaining at 0x1), I would expect the watchpoint to trigger when the variable is 1 with the bit being set. Instead, the watchpoint triggers on every bit toggle regardless of the value of the variable. It is unclear what the expected impact of these fields are and if they are working correctly.

Attempting to launch a debug session for an MSPM0L1307 LaunchPad using the XDS110ISO-EVM debug connection. Something is wrong with the connection and a dialog appears asking if a Force Reset should be performed. It has 2 buttons, OK and Force Reset. If the reset does not work the dialog is shown again. Clicking ok, X, or Force Reset just results in the dialog. being presented again. Even terminating the debug session does not exit this loop. Only way to clear it is to close CCS.

Calling session.breakpoint.add() with an undefined argument does not throw an error. On the contrary it successfully made a breakpoint at address 0x00000000.

For the MSPM0 device, the CS_DAP has many 'Unknown View' entries in the memory page drop down.

Using Ubuntu 22 in dark mode, the installer text is not readable, it works fine in light mode.

The registers view does not have support for a float format type. Both "Real" and "Scientific" are for decimal. It would make more sense to have those for float like CCS Eclipse.

When double clicking an element, there is no consistent behaviour between watch and variable views, as well as child/parent elements. They should all edit the value of the expression/variable.

Debug launch stuck with "Reconnecting to TI Cloud Agent" when multiple XDS110 based targets are connected. It doesn't seem to happen with all XDS110 based targets but just particular combination. It can be consistently reproduced with one LP-MSPM0G3507 and one LP-EM-CC2745R10-Q1 (via LP-XDS110). The latter board is likely the key factor. The workaround is to disconnect all the targets but one, then launch the CCS IDE. Then connect the other boards.

The default is fixed to 32-bit signed. Set default DSP data type based on graph expression

Debug launch fails if CCXML is in a path with unicode characters.

If a project has a linked resource such as a source file. If you create a folder in your project and drag that linked resource into it the physical file is moved in the file-system from its original location into the folder in your project. In many cases these files are part of the SDK so doing this will break any SDK projects that reference those files. Instead it would be expected that the link would just move into the folder and still point to the file in its original location.

When debugging code, the Watch windows don't show the correct values for bitfields, when bitfields are updated. The bitfield values are correct when the structures are first initialized, but as they are updated, changed bietfields update to no longer show the enum value, but instead to display the negative value of the enum typedef (such as -1 or -4).

In some scenarios, there is not any source available in the disassembly view. The PC target does halted at the main function in the editor.

Debug launch fails if CCS is installed in a path with unicode characters. It only appears to impact Windows and Linux. The Windows Unicode UTF-8 setting will workaround the issue on Windows. A workaround is currently unknown on Linux

CCS 20.3.1 incremental update breaks editor functionality such syntax highlighting, code navigation and most other advanced editor features. Clearing the various CCS cache does not help. The issue only impacts an incremental update to CCS 20.3.1. A full CCS 20.3.1 installation from scratch does not have this issue.

Loading .mjs files give an error in the scripting console. This affects both scripting console and target initialization scripts.

There appears to be a hard limit of ten executable actions that is available from the project "Executable Actions" context menu. There is not any hard limit in the project properties, however any after ten will not be accessible outside the properties.

ROV does not show data if it is opened while debug session is starting. # Start debug session (debug freertos project) # While session is starting, open a new ROV # Observe that once session is started, ROV is not updated to show the data # ROV needs to be reopened to see the correct data.

Selected line moves by 1 when stepping in disassembly view.

Existing address watchpoints don't enable in the UI when we connect to target Steps to reproduce: -> Start a debug session and set a watchpoint at an address. -> Terminate debug session. -> Start a projectless debug session. The breakpoint would be unverified. This is expected. -> Connect to target. This sets the watchpoint on the target, but it still shows as unverified in the IDE UI.

After changing the memory read/write behavior to use the status buffer, the sw bp code now ignores memory access errors because it provides a status buffer but doesn't check it.

The debugger is not aware of hardware breakpoint types until the first connect (when CTools is fully initialized). If symbols are loaded prior to the first connect, a symbolic breakpoint, like the one at C$$EXIT, will not be converted to a hardware breakpoint since the debugger is not yet aware of their existence. This can result in a software breakpoint being erroneously used when the debug state is later restored.

The simple workaround is to refresh the view.   The root cause is the late registration of refresh event handlers in {{{}CCSPinnableWidget{}}}. This was not a visible issue when we relied on the {{targetState.changed}} event for triggering debug views to populate with data. However, we moved away from relying on this event, as doing so was causing other issues. The {{refresh}} event also happens to be the appropriate event to rely on here.   The solution to this issue is to always call {{refreshTargetContent}} on the start of a debug session.

Debugger full verification is not correctly identifying mismatch between the data in the target memory and actual data in the loaded file. This issue doesn't happen in all cases, only with some specific files. The behavior varies depending on the CCS version: CCS 20.4.0: Incorrect CCS 12.8.1: Incorrect CCS 9.3.0: Correct There appears to be no issue with verification done by the flash programmer.

Today CCS will let you install on older macOS versions that are not supported. These may or may not work. The installer should detect that these versions are not supported and warn them that the product may not operate properly. In the case of macOS 10.15 (Catalina) we should block installation as it will not work as the Theia IDE prevents execution.

The Theia we include in CCSv20.4.0 detects if you are running on Catalina and blocks execution. This can be a problem for users who are updating an existing CCS installation. For example you can install CCSv20.3.0 on Catalina and it will run fine even though it is not supported. However if you then update to 20.4.0 your installation will be broken.

It appears that the $(GEL_file_dir) macro can only be used once for a function call. For example: GEL_TextOut("$(GEL_file_dir)/file1.c $(GEL_file_dir)/file2.c\n"); Would result in the GEL output: CORTEX_M0P: GEL Output: C:\ti\ccs1281\ccs\ccs_base\emulation\gel/file1.c $(GEL_file_dir)/file2.c The issue occurs with both CCS 12.8.1 and CCS 20.4.0.

Disassembly view sometimes only loads with one or two lines of disassembly. Resizing or collapsing and re-expanding the widget immediately fixes the issue.

CIO file handling fread/fwrite works ok when using load program, and in such cases, the files are found relative to the location of the loaded program executable. The same should happen when using load symbols, but the files cannot be found.

CCS doesn't set breakpoint in mapped source file. Workarounds include involve typing out the location of the breakpoint manually (either in the scripting console, or in a breakpoint creation dialog). In both cases, the breakpoint is set on the target but it does not appear in the source view.

Motorola S5 record type is not supported. Hence files with this record type can trigger an undefined record type error. According to the documentation, it is an option record for "count" which is typically not used. We can ignore it instead of returning an error.

Link step fails when .syscfg file and SysConfig-generated files are placed in linked folder.

When building the project, the IDE hang and turn into a blank screen. No issues building via CLI. Also, setting project build verbosity to "Super quiet" builds without issue in the IDE.

For a simple recursive function, after stepping into the function a couple of times, each subsequent call doesn't seem to be increasing the number of stack frames returned by the DS.

The build time for large projects with many source files and build options can be greatly impacted by various security software running on the machine. Some optimizations can be done within CCS to help alleviate this issue. # Build-output verbosity affects build-time. For rule commands, gmake writes each command into a temporary .bat file, then runs that .bat file. Switching to 'super-quiet' removes a few 'echo' commands for every compiled file. Today, build-output verbosity can only be set in workspace preferences. It should be allowed to also implement support for setting it per project. Also, support the ability yo add a new 'automatic' build-verbosity mode, which would auto-reduce verbosity based on project size. This 'automatic' mode should be the default. # Long compile commands affect build-time. CCS currently handles long compile commands by generating a few additional commands for each compile rule - these commands write all flags into a temporary .opt file, then run the compiler with that .opt file. This is a bit inefficient because all this is done at build-time, for each compiled file. This can be improved by pre-checking just the compiler include-path, and writing only the include-path into a separate .opt file during makefile-generation.

If user adds more HW breakpoint than are available on a particular device, the BPs are added, but in a disabled (queued) state. If user then removes one of the HW breakpoint, the queued BP is automatically added by the BP manager, but it shows as if it is added as a SW breakpoint, and also without a corresponding CORE attached.  It seems that the BP is correctly added as a HW breakpoint, but this is not confirmed. NOTE : This breakpoint also cannot be removed by clicking on it in the disassembly view and can only be removed in the breakpoint manager view.

The options "Watch" and "Number Format" only works for registers, not groups or bitfields. Should hide/disable the options in the menu.

Disassembly view does not show breakpoints in UI in some cases. One example is when debug session is in progress, closing and re-opeing disassembly view will cause the UI for breakpoints to stop working.