#define not marked as a definition

[Fomys]

Hello,

I believe there is an issue with the #define parsing in the code here. This definition is not being detected when searching for pm_ptr.

Additionally, just above it, pm_sleep_ptr is not even considered as an identifier, so it cannot be found anywhere.

Same for mmSPI_CONFIG_CNTL_1, at least used here and defined here (BTW on the same page you can see many missing links)

It is also the case for older linux version, I just checked with 6.3 and 5.19, pm_ptr and pm_sleep_ptr are not parsed correctly.

[fstachura]

It seems that pm_sleep and pm_sleep_ptr are filtered out by this line https://github.com/bootlin/elixir/blob/master/script.sh#L169

root@01eac2eb3583:/usr/local/elixir# ctags -x --kinds-c=+p+x --extras='-{anonymous}' pm.h | grep pm_ptr                                 
pm_ptr           macro       475 pm.h             #define pm_ptr(_ptr) PTR_IF(IS_ENABLED(CONFIG_PM), (_ptr))
root@01eac2eb3583:/usr/local/elixir# ctags -x --kinds-c=+p+x --extras='-{anonymous}' pm.h | grep pm_sleep_ptr
pm_sleep_ptr     macro       476 pm.h             #define pm_sleep_ptr(_ptr) PTR_IF(IS_ENABLED(CONFIG_PM_SLEEP), (_ptr))

Notice CONFIG_ in both lines. Same with mmSPI_CONFIG_CNTL_1 and others from that page. All contain CONFIG_.

[tleb]

One line fix, nice catch @fstachura and @Fomys:

... | grep -avE "^operator |^CONFIG_" | ...

I'll send the fix. Maybe using -e twice if that is supported with the current flags passed.

[fstachura]

@tleb I have a fix ready, the question is - does filtering identifies that start with CONFIG_ make sense? There are some that are not in Kconfig

[tleb]

That sounds like a rather unrelated question. Indeed I don't see a reason for filtering those out; they are defines nonetheless that might interest someone someday. Sadly Git log cannot help us: 72571fb.

Attaching the output of the following commands:

⟩ fd -tf | grep -v Kconfig | \
	parallel -L1000 ctags -x 2>/dev/null | \
	grep ^CONFIG_ | sort > /tmp/linux-config-c.txt
⟩ fd -tf | grep -v -e Kconfig -e '\.c$' -e '\.h$' | \
	parallel -L1000 ctags -x 2>/dev/null | \
	grep ^CONFIG_ | sort > /tmp/linux-config-remainder.txt

linux-config-c.txt
linux-config-remainder.txt

For the operator<space> grep, the matches are:

⟩ fd -tf | parallel -L1000 ctags -x 2>/dev/null | grep '^operator '
operator         member      174 kernel/trace/trace_events_hist.c enum field_op_id operator;
operator ->      function     48 tools/testing/selftests/bpf/test_cpp.cpp const T* operator->() const { return skel; }
operator ->      function     50 tools/testing/selftests/bpf/test_cpp.cpp T* operator->() { return skel; }

Let's do that in two steps: first fix the current issue (at least in the code, we don't have to do the full reindex straight away). Then we can discuss about that indexing logic.

---

edit: related:

elixir/update.py

Lines 318 to 326 in 55921f1

	if (db.defs.exists(tok) and
	not ( (idx*idx_key_mod + line_num) in defs_idxes and
	defs_idxes[idx*idx_key_mod + line_num] == tok ) and
	(family != 'M' or tok.startswith(b'CONFIG_'))):
	# We only index CONFIG_??? in makefiles
	if tok in idents:
	idents[tok] += ',' + str(line_num)
	else:
	idents[tok] = str(line_num)

[fstachura]

@tleb I think #344 can be merged? It works as a quick fix for this.
Later this filter should be moved to a project specific part of Elixir, so that it's only ran on Linux (and perhaps other projects that use Kconfig with CONFIG_ prefix).
And also, functions that start with operator_ should only be filtered out in C++.