LLVM Tutorial #10: Getting Out of Trouble

In this series I walkthrough the LLVM "Kaleidoscope" Tutorial, where you follow step by step to create your first programming language frontend using LLVM as the backend. Last time we finished writing code to generate LLVM IR code, but we hit a snag. In this episode I put my debugging skills to work and get out of trouble.

Transcript

The following transcript was automatically generated by an algorithm.

00:00 : all right we're back so um I'm going to
00:04 : continue this series where we follow the
00:08 : LLVM tutorial that's on LLVM dot-org
00:13 : we're at the end of chapter 3 and near
00:17 : the excuse me near the end of the last
00:21 : episode we hit a snag
00:24 : we got the program to compile which was
00:29 : not easy but uh once we got that to
00:31 : compile we ran it and the moment we
00:33 : throw something at throw something at it
00:40 : it's sick folks so you're gonna need to
00:43 : debug this I do want to show people to
00:49 : you that you or anyone watching how to
00:54 : debug C++ code I think that's very
00:58 : important to be competent in debugging
01:03 : your programs one of the faults in my
01:06 : opinion of C++ is when when something
01:10 : wrong happens during runtime it doesn't
01:14 : give you any message that's helpful to
01:17 : you so then it makes your investigation
01:21 : more difficult but but but that's kind
01:26 : of you know we just have to deal with
01:30 : that so when you have a segmentation
01:32 : fault a lot of times that's because a
01:36 : reference reference in your code so
01:43 : there's a couple of approaches we can
01:45 : take one is use a step debugger to step
01:49 : through the code line by line and figure
01:51 : out which line of code the problem
01:55 : occurred another one is you put in some
01:59 : print statements sprinkle some print
02:03 : statements in the program and then try
02:06 : to gather information from
02:08 : from from those print statements as you
02:10 : execute the program and then another
02:13 : approach would be let's simplified
02:15 : program so that the testing path is
02:22 : easier and more repeatable I think what
02:26 : I'm gonna do probably a combination of a
02:29 : couple of things first I'm gonna
02:30 : simplify this program instead of doing
02:33 : this main loop thing and then I have to
02:36 : go through a series of user actions
02:40 : before I can test something else I'm
02:44 : going to simplify it so that is running
02:47 : the main loop I'm just going to let me
02:53 : think
02:55 : currently it just reads from the body I
03:05 : think I might be able to do is just
03:09 : parse top-level expression
03:19 : it's the only thing in the program
03:27 : whoopsies it's the only thing in this
03:31 : program that's supposed to be Auto with
03:34 : a name and then I guess if that didn't
03:39 : work out if that didn't work out I'll
03:43 : say didn't work no parse sound but if it
03:55 : did give us something then we'll do this
03:59 : print out the ast print with eros which
04:04 : I don't really understand what the Paris
04:06 : thing is doing I'm just copying the code
04:08 : from here oh no we need to Cochin first
04:14 : whoopsies given the ast we need to call
04:21 : the code ran on the ast node to get the
04:25 : ir representation and then it's actually
04:28 : the ir representation from which we're
04:31 : gonna print so we're gonna try this run
04:35 : this program but I think I can run this
04:39 : program type in something hit enter and
04:45 : see there is up or I can type in a
04:49 : program saving a file and then type the
04:53 : file into this program so let me rebuild
04:57 : this program huh I have a build script
05:01 : that I can just run to rebuild it
05:13 : okay so I think I'm just gonna try doing
05:18 : it from the command line anon token when
05:23 : expecting an expression already illegal
05:29 : instruction for whoa not expecting that
05:36 : at all legal instruction for illegal
05:45 : instruction for may be unknown token
05:59 : expect it got here interesting something
06:07 : I'm trying to do what I'm gonna try to
06:09 : do is make a file and pipe it in so I'll
06:13 : [Music]
06:16 : follow that kaleidoscope Clara scope KS
06:28 : let's say the the first program is just
06:31 : one plus one I think that should be a
06:35 : valid expression so I'm going to cat and
06:38 : you next cap will print out the file and
06:43 : I'm gonna pipe that to that program and
06:48 : we still get this error here so I think
06:54 : we need to debug this let's just go into
06:57 : this DEP debugger so in order to use
07:03 : this step debugger in to make sure we
07:06 : built the program with debug mode on so
07:13 : so I'm gonna remove - you put the two
07:17 : comments on the top so I do have debug
07:21 : mode turned on so we're good - G means
07:24 : debug mode on
07:26 : I think this optimization might make it
07:30 : a little harder maybe but I'm gonna try
07:32 : it and if it's still hard I might lower
07:38 : the optimize level to 0 or 1 you know
07:43 : I'm gonna how do i pipe stuff into it
07:49 : don't know not piping input into it
07:58 : maybe I'll just debug it like that so
08:00 : I'll use LOD B to debug the program
08:04 : another module in the LLVM
08:07 : infrastructure I have you been using lol
08:10 : DB even just for coding and C and C++
08:16 : so I'm quite familiar with how to use it
08:20 : I'm gonna set a breakpoint
08:23 : [Music]
08:25 : I'll find a line number I'm gonna say
08:28 : line 565 so I would say B I have to type
08:33 : in the name of my
08:35 : file and then I'll save 500 and what was
08:39 : it five hundred and sixty-five and it
08:41 : set a breakpoint and I'm gonna use the
08:44 : arc command to run the program and it
08:47 : broke here I set a breakpoint here I'm
08:52 : going to go and go into GUI mode I would
08:55 : set the target stopped okay
09:04 : it skipped over the line that I wanted
09:07 : to stop I wanted to stop here
09:14 : interesting I thought it would have
09:16 : error it out already because the the
09:21 : error came from a departure this is
09:25 : quite surprising to me actually so what
09:32 : is the FN AST let me look at that so you
09:36 : can actually examine the variables and
09:38 : down here although this is looking kind
09:42 : of gnarly very very dense so FN AST is a
09:48 : unique pointer which contains a
09:54 : compressed pair
10:01 : okay within it it's got a proto ast in
10:09 : it because another unique pointer and
10:13 : has these compressed pears in it this is
10:18 : actually harder than I thought to
10:20 : navigate because this these data
10:22 : structures is that we're using from the
10:25 : standard library is pretty complex in
10:28 : structure help finally we get to the
10:35 : name and probably the value I guess I'm
10:47 : gonna try to just step and we hit an
10:52 : error 'invalid up when I try to step
10:58 : over the Kochan function so I'm gonna
11:01 : restart I kind of want to verify that in
11:07 : fact I don't understand why it didn't
11:14 : pause on line 565 actually that was
11:19 : confusing I'm gonna try 564 just in case
11:24 : and it still stopped here okay I'm gonna
11:28 : step in stepping in gets me the air so
11:32 : right now I'm thinking I should
11:34 : recompile it with a lower optimized
11:38 : level and I think I have a better
11:43 : debugging experience if I did that so
11:46 : let me rebuild it
11:47 : I changed the optimization level two to
11:50 : zero so it won't try at all to do any
11:53 : optimization
12:04 : do that again now I set my break point
12:08 : I'll do it 565 run it and actually it
12:13 : stops at the right line
12:15 : thanks okay so now we're at this parse
12:19 : line and I'm going to step into the
12:20 : parser there - I actually I expect this
12:28 : not to work because it's trying to reach
12:32 : characters from standard input and we
12:37 : don't have any standard input kind of
12:39 : setup so I actually wonder what it's
12:42 : going to do
12:49 : what are we parsing a top level
12:53 : expression yeah we are parsing the top
12:56 : level expression which is just gonna
12:59 : purse a call parts expression so what is
13:04 : Kurt car took what is the current token
13:09 : and current token is zero it looks like
13:16 : is that right I'll step and so unknown
13:21 : token when expecting an expression
13:24 : current Kurt okay wait
13:40 : did anyone even call get Tok get next
13:47 : token cuz if you gonna call parse
13:54 : expression there's parse definition
14:05 : parse top level expert doesn't call get
14:09 : token surprisingly I thought that it
14:12 : would and parse expression doesn't do it
14:17 : either that might be a problem and then
14:22 : parse primary also doesn't call get next
14:25 : token so there's no one to call get next
14:29 : token oh wait this just expects the
14:35 : current token to already be an
14:38 : identifier so I think that might be the
14:44 : problem or at least one of the problems
14:46 : so I think what we need to do and I'll
14:51 : look at their code I'm suspecting that
14:55 : in their tars top level it probably I
15:05 : don't see it the parse expression I
15:11 : don't see it here either
15:17 : and oh maybe oh oh oh the main loop oh I
15:21 : see the main loop does I get the handle
15:25 : definition no I don't get it who is
15:36 : calling to get next token maybe we need
15:46 : to call get next token manually first oh
15:49 : oh oh that's initially there's just one
15:54 : car to get next token okay I'm just
15:58 : gonna put that here okay let's try it
16:06 : again I think is something definitely
16:09 : needed to happen
16:22 : okay I'm just gonna try it anon debug
16:26 : mode and it seems like it's waiting for
16:28 : me to type in something now which is
16:30 : good so I'm gonna type in one plus one
16:36 : it didn't seem to do anything I think I
16:44 : might do a printout here first
16:55 : ast if it ever finished parsing DST
17:09 : there's an end of file character at all
17:17 : no it didn't and I would just say parsed
17:21 : est but then after that there was a sick
17:24 : fault
17:25 : first ast maybe showed this point
17:54 : one plus one I did person est it could
18:06 : be the Cochin that went wrong but how
18:13 : can we run this in the debugger that's a
18:17 : question we might be able to let me try
18:21 : it
18:22 : hello DB my let me set a breakpoint at
18:31 : line 5 67
18:35 : that's this line cuz I want to step into
18:39 : the Cochin I suspect the Cochin is
18:41 : actually the code that calls in the
18:46 : error so 567 and then I'm gonna run it
18:54 : and I think this allows me to type in oh
19:03 : you know what maybe what I can do is
19:12 : have a get the parser to have a special
19:16 : character that tells it to stop maybe
19:22 : that might require a bit more work to do
19:30 : all right oh we did stop also all right
19:34 : I mean drop into my GUI that worked okay
19:38 : I'm stepping into the Cochin yeah okay
19:41 : we're in business now
19:42 : I think this dis these variables are
19:47 : still kind of hard to observe because
19:49 : they're wrapped within these unique
19:52 : point or structures which are kind
19:57 : have a complicated structure but then
20:02 : well is it C so in so we're actually
20:08 : calling Cochin on a function est oh that
20:13 : actually makes sense because um the
20:16 : top-level parser always wraps the
20:19 : expression in this top-level anonymous
20:22 : function so yeah it parses the
20:26 : expression but it wraps it in this
20:29 : anonymous function that doesn't have a
20:31 : name so that's why it's the top-level
20:35 : ast know is actually a function so we're
20:39 : gonna get a function it looks like
20:44 : that's the point where it failed is that
20:51 : true yeah and gave me the machine
20:56 : instruction I'm not really good with
21:01 : machine code but this means move Q it
21:05 : moves it is moves the variable inside
21:11 : the address that it's contained in this
21:15 : register into that register that's kind
21:19 : of what that's saying and I guess I
21:21 : could look at this code but you may not
21:25 : be able to make too much sense out of it
21:27 : but I guess what I gather based on that
21:32 : is that the line that actually caused
21:37 : the segmentation fault is the functions
21:43 : [Music]
21:51 : it's not in the parser but it's in the
21:54 : code Chen it's in the functions Cochin I
22:00 : think it's the function definitions
22:02 : cogent method where is that where is
22:06 : that I'm trying to find it this is in
22:12 : the parser though the code gen stuff is
22:16 : higher up okay yes it was this one maybe
22:23 : it's because proto is undefined that's
22:27 : one possibility
22:27 : another possibility is maybe it was
22:30 : because the module was undefined so
22:33 : let's see which one it is you could
22:41 : actually print it out but we could also
22:43 : just go back into the debugger not sure
22:46 : which one is faster honestly I'm just
22:49 : going back into the debugger arrow a few
22:53 : times to get my brick point back run the
22:55 : program type in I'm just gonna type in
22:58 : one actually instead of one plus one to
23:01 : simplify it and then now we're in here
23:04 : go into GUI mode because I do like
23:07 : cooling mode let's look at now what is
23:11 : the value of the module we'll have D
23:17 : context might need to make the text
23:21 : smaller so I can see more stuff let me
23:25 : try it again I making the text smaller
23:31 : so I can see you have more real estate
23:35 : in the debugger okay going into the GUI
23:41 : going into the code 10 function all
23:43 : right now I want to look at the module
23:48 : where is the module have builder of this
23:52 : map
23:55 : Kurt we'll have Kurt oak da madro okay
24:00 : we have done module it's actually got
24:02 : stuff in it okay it could be the proto
24:06 : is not defined Oh proto that proto arrow
24:14 : get name did we ever write the get name
24:18 : function maybe that's what it is
24:21 : what is proto do we have a proto proto a
24:30 : proto is it is an instance variable
24:34 : within the function so it would be under
24:37 : this a proto does have something in it
24:41 : it it's a unique pointer okay that
24:46 : points to something that has a name but
24:48 : but I don't think I implemented the
24:50 : getname method for proto honestly and
24:54 : that might be the reason there's a
24:58 : getname method defined but don't
25:01 : remember implementing it and in their
25:06 : code they probably have it implemented
25:18 : know what
25:21 : oh oh they just inlined it they just
25:28 : inline this to return name did I do that
25:33 : yeah I did that too
25:48 : mine has it reversed does that matter I
25:53 : don't think it matters but also theirs
25:57 : theirs doesn't have to work virtual
26:01 : before before this I don't think that
26:06 : matters either but you never know I
26:09 : guess
26:12 : how is mine different there's this work
26:26 : though I have tested compiling their
26:29 : code and running there too and there
26:31 : does work and if you feel like skipping
26:39 : this part and just going to that part
26:41 : yeah by all means I'll show a working
26:45 : version in the next video and for this
26:50 : video I'm just gonna spend the time
26:52 : trying to debug my version mm-hmm
26:57 : so getname I could try to yeah I could
27:03 : try to split it up and yeah let me try
27:07 : this
27:11 : I'm going to not in here but in here I'm
27:19 : gonna get the name into a separate
27:21 : variable Prudhoe arrow get name got the
27:28 : name so as a second step I was a
27:34 : printout like defining function name and
27:46 : then I'll say successfully got from
28:03 : wanderer from the module
28:19 : alright gonna rebuild it and then run it
28:22 : [Music]
28:27 : okay good it's a defining function Mt
28:47 : which makes sense because it is supposed
28:51 : to be an anonymous function but calling
28:54 : the model yet function so is it the mod
29:00 : Road that get function it might be the
29:04 : model that get function because the mod
29:07 : rope know the module is a unique pointer
29:10 : actual but the module get function did
29:20 : not work for some reason so I'm gonna
29:27 : isolate it to this so I'm just gonna
29:30 : have the program only do that I'm gonna
29:39 : skip the whole parsing process and just
29:42 : isolate - oh come on
29:48 : why is my editor behaving this way I'm
29:56 : gonna try to get the function
30:03 : like that and see that sick folks I'm
30:06 : gonna print out something build it again
30:17 : and see if it's sick Falls hmm it does
30:30 : sick for so that's actually a great
30:32 : result even though you might not think
30:35 : that because there probably means
30:39 : there's something about oh we need to do
30:45 : this we do not instantiate the module
30:50 : holy crap that that was pretty important
30:56 : yeah it turns out we needed to
30:58 : instantiate the module and we're gonna
31:04 : call it the module my cool
31:07 : apparently okay did they tell me to do
31:14 : that come on man
31:17 : I thought this was a step-by-step follow
31:22 : along driver code so it's physically
31:27 : saying hey you know what just copy my
31:29 : code yeah it's fine okay so I think that
31:35 : is the line I was missing let me try it
31:38 : now
31:47 : okay now I'm in the 6s okay beautiful
31:53 : okay so that was the line I was missing
31:56 : and now I'm going to try this my
32:01 : previous snippet oh come on
32:04 : yes code I have this autosave feature on
32:09 : and I swear I'm gonna turn it off very
32:12 : very soon okay
32:16 : it's really bothering me right now okay
32:23 : try to type some code in and then it's
32:27 : going to if we're lucky
32:31 : oh my god that's freaking awesome
32:41 : so basically I typed in a 1 my program
32:47 : the text of my program my entire program
32:49 : is one and then he translated that
32:55 : program in the Kaleidoscope language it
32:59 : translated that program in the language
33:02 : of LLVM ir and the code in LLVM ir looks
33:09 : exactly like that and it's saying i'm
33:13 : gonna define a function that returns a
33:16 : double code the function at zero for
33:20 : some reason and it has zero arguments
33:24 : and it has a body which is called entry
33:28 : and it's going to return a double and
33:32 : the value it is gonna return this one so
33:36 : that worked and I'm gonna revert this
33:41 : code back to this main loop and play
33:44 : with it a little bit
33:46 : and then in the next and out stop this
33:51 : video and then in the next video I will
33:53 : play with play with the repple a little
33:58 : bit so I'm gonna need to instantiate the
34:05 : module first as it turns out and then
34:12 : I'm gonna run the main loop okay okay so
34:28 : that's the repple and you make the text
34:31 : bigger again so we already did the one
34:35 : okay so one gives us this program
34:40 : successfully got point maybe I should
34:43 : remove define double read top level
34:51 : expression where does it say read top
34:54 : level expression okay maybe I do a new
34:59 : line here
35:01 : read read top level this is red X turn
35:07 : red function definition and I'll get rid
35:10 : of my extra print statements
35:19 : I'm gonna recompile the program and then
35:24 : dude to the to the ripple one more time
35:32 : okay
35:34 : one plus one semicolon and then Oh
35:38 : actually has a smartly pre added to one
35:43 : plus one okay
35:45 : so okay great the rep who works I'm
35:48 : gonna stop this video here in the next
35:49 : one we'll play some more with the rebel