TODO

1. nested library tracing (done)

• We can see meaningful enough library functions without user functions.
• We don't miss some important library calls.
• We can remove the restriction of recompiling with -pg or -finstrument-functions.
• done: #141
• merged: c618ccc

2. (python) script enhancement (done)

• arguments / retval support for python #144
• option handling in python script (in order to make a given script self-contained)
• script filter (e.g. @script= ) or in-script filter
• merged: 66d5e54

2-2. remaining (python) script enhancement

• drawing various charts using matplotlib.
  - drawing malloc and free to see memory allocation graph
  - drawing RSS charts by using read trigger
  - many others...
• event handing in (python) script
• collecting useful scripts
  - such as tools in https://github.com/brendangregg/perf-tools

3. automatic argument / retval display for known libc functions

• We can do this for some known functions such as malloc and free.
• Some functions that use char* string such as strcpy and execv can show eye-catch string output.
• We can support some functions at the beginning and incrementally add one by one.
• As a first step, we can simply provide an option file that contains a list of -A and -R options for specific functions.
• We can also provide library's return value checker. (such as if malloc returns NULL in some cases)
• almost done #158

4. make filtering more convenient

• currently regex and glob matching is used in a mix
• support to select regex, glob or simple substring match based on user's chioce
• consolidate similar options so users don't need to type function name multiple times

5. return-time trigger

• current trigger action is called only at function entry
• reuse -R option to allow it called at function exit

6. call graph diff

• We can record twice and compare the summary using report --diff as of now.
• We can also compare the call graph diff to identify the specific difference in details.
• We may provide another GUI output for showing the diff efficiently.

7. builtin chrome trace viewer

• Currently, showing chrome trace output requires many steps.
  - uftrace record -> uftrace dump --chrome -> goto chrome://tracing -> press Load to open file
• I would like to make it as the following steps
  - uftrace record -> uftrace chrome -> it shows a URL link -> goto the link in any other remote machine.
• It requires integrating chrome trace viewer inside uftrace repo.
• It also requires to write a simple node script to provide a web link.
  - The following node program can simply host html trace file on browser
  - https://stackoverflow.com/a/8427954/921491
• It's mainly to make it simple to use for general users.

8. DWARF debug info support

• it can be used to specify argument and return value automatically
• it provides source location info (file:line)
• it might be used as a new filter option - (e.g. filter all functions in this file)

9. user interface improvement

• current console based UI lacks user interaction
• maybe (ncurses-based) TUI can be added
• I have no idea about GUI (but welcome contribution!!!)

10. handling demon programs

• We need to implement a way to trigger tracing on and off using signal or socket.
• I prefer implementing in socket so that it can be used in embedded system.
• It can be used with recv command.
• We can trace multiple processes with data file merge feature.

11. full dynamic tracing

• compiler support free tracing without recompilation.
• It requires to decode each instruction and emulate.

12. more C++ std::* support

• understanding STL containers
• std::string (done!)
• std::vector, std::list, std::shared_ptr, std::unique_ptr, etc...

13. attaching to already running process

• dl-library linked or not?

14. other language support

• Rust or Go ?
• Python - $ python -mtrace --trace --timing ./runtest.py 161 with python interpreter itself.
• JavaScript (need to modify engine itself)

etc.

• automatic color for each context (kernel, library, event)
• --help with pager
• more auto completion
• value filter