[tor-bugs] #19180 [Core Tor/Tor]: Add new compiler warnings

[Tor Bug Tracker & Wiki]

#19180: Add new compiler warnings
--------------------------+------------------------------
 Reporter:  nickm         |          Owner:  nickm
     Type:  defect        |         Status:  needs_review
 Priority:  Medium        |      Milestone:  Tor: 0.2.???
Component:  Core Tor/Tor  |        Version:
 Severity:  Normal        |     Resolution:
 Keywords:  029-proposed  |  Actual Points:
Parent ID:                |         Points:
 Reviewer:                |        Sponsor:
--------------------------+------------------------------

Comment (by nickm):

 ok, I really think I'm done for now.  I've gone over the easy and not-too-
 hard cases.  Of the warnings I originally considered, they are either Out,
 Done, or not-yet-done.

 Here are the ones that I thought we might want to do where I decided
 against them:
 {{{
   suggest-attrubte=const (4.6)
   suggest-attribute=pure (4.6)

     Rationale: these just suggest attributes that the compiler can use
     to optimize code better (if we get them right) but which will make
     the code incorrect (if we're wrong).

   jump-misses-init (4.6)

     Rationale: triggers all over.  In many (all?) cases, the missing
     initializer is for a variable that is not used after the jump. Our
     existing static analysis tools SHOULD catch the cases where we can
     _use_ an uninitialized variable.

   unsuffixed-float-constants (4.6)
     Not even like a bug; 1.0 is a fine way to spell the (double) 1.0.

   strict-aliasing=5??? (3)
     Requires strict-aliasing, which we disable.

   disabled-macro-expansion (clang)
     Triggers in stdio.h

   extended-offsetof (clang)
     We require this extension.

   used-but-marked-unused (clang)
     We need to be able to use "unused" to mean "maybe unused".
 }}}

 Here are the ones that we might want to look at later:
 {{{
   cast-qual (4.6)

     Rationale: triggers everywhere, even in some pretty normal C.  Would
     be nice to have it trigger less, but would need to blow up a bunch
     of API things.  Bigger project.

   conversion (4.6)

     Rationale: triggers all over.  Probably wrong code in some
     cases, but careful thought needed in most Bigger project.

   sign-conversion (4.6)

     Triggers ALL OVER.  Quite possibly a bug in some cases, though.
     Bigger project.

   cast-align (3)
     We already do this safely. Need to re-test on a system with
     stronger-than-intel alignment rules, though.

   shadow (3)
     mistake; worth fixing.

   switch-default (3)
    Not sure this is a good idea; somet of these look like mistakes,
    but some don't.

   assign-enum (clang)
    triggers all over; worth fixing.

   conditional-uninitialized (clang)
    triggers all over; not sure whether this is worth fixing.
 }}}

 These should not become on-by-default, but they're worth hand-auditing
 {{{

   strict-overflow=3...5 (4.2)

     Behaves pretty differently on different GCC versions.

     We get warnings for just about every case where we have pointer
     math in an addition. That seems nutty.

   padded (3)
     Not a mistake.  Worth looking over for hand-audit purposes, but mostly
     harmless.

   unsafe-loop-optimizations (4.1)
     Worth hand-auditing, but triggers on every kind of interesting for
 loop.

   covered-switch-default
     Usually this is defensive programming, but it could be a mistake
     in some cases, or could cover up future mistakes?
 }}}

 Aaand the bug19180_easy branch is still needs_review. :)

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/19180#comment:10>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online