The
kind of software being written materially influences the semantics and use cases for TODO. I think much of the disagreement in this comment thread is based on a tacit assumption that everyone works on the same kinds of code bases. A web app is not the Linux kernel, and the priorities and concerns in development when annotating code will differ.
For example, in much of the database code I work with, TODO is used for optimization banking. In performance critical applications, this is the practice of pervasively tagging (often with TODO) micro-optimization opportunities in the code. When a bug fix or new feature unavoidably creates a significant performance regression (I use 5%), you implement several of the "banked" optimizations documented as TODO in order to bring performance back to parity. For this reason, a TODO may linger in the code base indefinitely and that is okay. It is a searchable annotation for code work that may never need to be done.
You do not put these kinds of TODO items in an issue tracker because they are extremely local and contextual to the code they apply to. Filling the issue tracker with thousands of these kinds of issues which can't be understood without looking at the specific lines of code the comment was applied to has high overhead with negligible value. You do open an issue in the tracker for performance regressions in a subsystem, or if you want to improve the performance by some metric, which directs a developer to start looking at the banked optimizations in the TODO comments inside the code.
Another rarely discussed use case for TODO is assumption verification around the internal behavior of external dependencies. Developers make assumptions about code dependencies (e.g. Linux syscalls or compiler code generation) based on trivially observable and testable behaviors. For code that needs to make strong assumptions about dependency behavior for robustness, it is frequently impractical at the time of writing code to verify that assumptions hold under all conditions or that they will remain true over time (see: Linux fsync() behavior). This becomes an issue that is never "resolveable" in a meaningful sense because it is a moving target and time-consuming to research for any particular case. The TODO is a reminder to re-verify behavior if anything has changed in the software environment or a new bug has shown up and usually notes what has been verified in the past.
Feature and bug work belongs in an issue tracker but there are many "issues", particularly ones that are non-functional and require extremely local context or can never be properly resolved, that are often best served in direct code annotation.