LLVM Tutorial #19: "If" It Can Work

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 wrote finished up writing the code for implementing the if statement, but hit a snag at the end. This time we go into full debug mode and try to get it working. Can it work? As a bonus, we also explain a little about the formulation of SSA form in the the output code, and why it isn't non-trivial in our case.

Transcript

The following transcript was automatically generated by an algorithm.

00:00 : hello there this is Toby and welcome
00:03 : back to the LLVM tutorial walkthrough we
00:08 : are at the end of the first half of
00:15 : chapter 5 where we are attempting we
00:19 : have attempted to implement if
00:22 : statements in the language the next
00:27 : section is going to be the for loop but
00:29 : before we go into the next section we we
00:33 : need to get our program working for us
00:35 : so I dutifully follow the instructions
00:38 : to to build out the parser the lexer the
00:43 : parser and the code generator for if
00:45 : statements but we doesn't seem to be
00:47 : working so I'm gonna do some debugging
00:50 : during this section so what I'm gonna do
00:57 : I think is because my first suspicion is
01:00 : that the parser is not working so I'm
01:04 : just gonna do some printouts I believe
01:09 : so I'll say I'll print some stuff to the
01:14 : console we'll find out the fact that the
01:22 : this part if expert function is start it
01:29 : let's see I will say
01:45 : otherwise I will say if we didn't get a
02:00 : expected done this log error should
02:03 : already print expected here will say
02:10 : done parts done success only comes here
02:19 : if we didn't get it done then I'll say
02:22 : horsten fail and then we're gonna push
02:28 : the else part we do this the same kind
02:32 : of behind earning poor success and then
02:43 : parts the whole thing first
02:46 : if expert success ok let's compile this
02:52 : and hopefully we'll get some more
02:55 : visibility and the parser or whether it
02:57 : parse the if statement correctly or not
02:59 : I guess one thing I could do is run
03:04 : their version and come up with a simple
03:08 : test case that works in their version
03:11 : and then I can use that same test case
03:13 : to test my version
03:25 : okay my program compiled ignore all the
03:31 : warnings so f1 then to l-3 does that
03:38 : work evaluated to 2 if 0 then 2 l's
03:43 : 3 3 looks good let me try to do that in
03:49 : my version that was their version their
03:52 : version elapsed the parser succeeded so
04:06 : maybe the problem isn't the parser that
04:10 : the problem might be with the code
04:15 : generator so I'm gonna do the same kind
04:19 : of logging in the code generator maybe
04:31 : the problem is that we don't have this
04:35 : switch statement for if statement could
04:38 : that be why I mean let me go to the
04:57 : kujan code the code gen code for this if
05:03 : statement see if it succeeded in
05:06 : generating the LPM code
05:18 : maybe I do a step debugger that's also
05:22 : possible and yeah you know what I'm
05:26 : gonna do a step debugger
05:32 : does that work yeah this to the step
05:38 : debugger so I'm gonna do ll DB which is
05:41 : the ll debugger LLVM ste bugger very
05:48 : good debugger I use it even when I'm not
05:51 : using when I'm not doing LLVM related
05:55 : projects so you used it oh l DB so I'm
06:03 : gonna set a breakpoint line 3:45 so say
06:09 : be my Lang CC 3:45 I'm gonna run that 1
06:24 : 2 3 hit pause there
06:29 : great I'm gonna go into GUI mode oh man
06:34 : what just happened why are the border
06:39 : written with X's and Q's this makes no
06:46 : sense
06:46 : okay I'm just gonna shut a step through
06:49 : this oh it couldn't even parse the
06:53 : condition oh we didn't know I mean not
06:58 : parse sorry this is not the parser is
07:00 : the coach and it couldn't even cogent
07:02 : the condition and return the null
07:05 : pointer
07:13 : why would that be so I mean look at what
07:19 : the code what the condition was that
07:25 : might be difficult there's a number X /
07:49 : ast it should have code tender and
07:56 : constant
08:19 : I can't look at their version of this
08:22 : code they have this override maybe it's
08:31 : important to have this overwriting
08:34 : because I don't have that I'm gonna try
08:39 : to keep that fixed it's really bizarre
08:51 : that the border are drawn with letters I
08:57 : don't know why that happened but let me
09:00 : recompile my program you know we do the
09:12 : same exact test with the debugger see
09:16 : then it'll actually give me a the code
09:19 : trend for the number expression will
09:23 : give me a constant value back instead of
09:25 : a null pointer
09:34 : it is quite strange that this problem
09:38 : has not surfaced before until now
09:41 : not sure why if t : : coat down 345 yes
09:59 : still 345 I'm gonna type if one then 203
10:10 : GUI mode still weird
10:15 : I'm gonna step in
10:26 : oh shoot
10:29 : that should have been a knot we did get
10:34 : a value back this over right I don't
10:38 : think matters I just didn't have this
10:43 : knot okay you know you make mistakes and
10:53 : that's how you learn folks yes I for
10:58 : some reason when I was copying this code
10:59 : I didn't put this exclamation point
11:02 : there you know that's why things are not
11:05 : working you know
11:19 : to cook to code to quote one of my
11:24 : friends Rachel her saying is if
11:30 : something seems too hard is probably
11:32 : something stupid and in this case it was
11:37 : was something stupid indeed that doesn't
11:44 : mean the person who wrote the code is
11:46 : stupid though stupid just means it was a
11:53 : elementary mistake
12:03 : okay see if it works now if one then two
12:10 : L's three yes if statements work now so
12:17 : I can do something like define a
12:28 : function okay all right I don't have I
12:41 : don't have strings in this language
12:51 : invalid binary operator we don't we
12:53 : don't have greater than sign yet of
12:56 : course so if X is less than five how
13:04 : about that then one zero so okay of ten
13:10 : gives us zero okay four gives us one so
13:16 : we have if statements implemented now
13:19 : great okay well thank you for watching
13:25 : this debugging session I'm trying to
13:29 : think is there anything else left to say
13:32 : about if statements and five functions
13:35 : [Music]
13:43 : I guess I'll just show you one more
13:46 : thing about ya
13:48 : the five function and the weirdness of
13:50 : the five function so before earlier in
13:56 : two episodes ago we gave this example of
14:03 : like if you have an arbitrary program
14:06 : with mutable variables so in the case of
14:09 : this X we assign values to it twice so
14:12 : it's immutable variable Y as well we
14:16 : have this branch that assigns to it and
14:18 : this branch also a science to it when we
14:21 : convert this program with mutable
14:24 : variables in it into SSA form then we
14:29 : have to do this stuff and that requires
14:31 : a five function I want to put what we've
14:36 : done into this same context so let me
14:44 : make another file entitled - so the
14:51 : reason that having to do this five
14:55 : function which was not quite as onerous
14:58 : as in the general case for the if
15:03 : expression implementation that we did
15:06 : for the Kaleidoscope language is the
15:11 : following in the Kaleidoscope language
15:17 : the Kaleidoscope language well currently
15:20 : first of all currently does not have
15:22 : mutable variables so if in kaleidoscope
15:26 : you do something like you know what I
15:29 : did something like this this is an if
15:45 : statement in kaleidoscope and if if
15:49 : statements in kaleidoscope are not if
15:51 : not like if statements in jobs
15:53 : or see if statements in 'caleidoscope
15:57 : are expressions as in this if statement
16:02 : is an expression that give you either a
16:09 : 1 or a 0 right depending on the
16:12 : branching it gives you a 1 or a 0 which
16:15 : means that in a sense it's like does it
16:20 : this temporary value the like you can
16:23 : assign this assign the value of the if
16:27 : statement well currently you can't
16:28 : because currently you
16:30 : we don't have variable assignment yet in
16:33 : kaleidoscope that's coming in an
16:35 : upcoming chapter but not yet but um but
16:40 : conceptually is like it's if statement
16:44 : yields a value whereas the if statement
16:48 : in conventional programming languages
16:50 : like JavaScript do not and because
16:54 : because in kaleidoscope and if stamen
16:57 : yields a value and therefore it has to
17:00 : you either like in in one branch it
17:03 : needs a certain value and another burns
17:05 : it uses a certain value it lends itself
17:08 : to be for us to be able to easily
17:12 : generate this five function because in
17:15 : essence it's basically saying a city
17:18 : output value of the F statement was Y
17:23 : then we're saying y1 is one why would
17:32 : meant to say y1 so let's write this in
17:35 : SSA form right we have an X let's call
17:39 : it X 1 if X 1 then Y 1 is equal to 1 and
17:44 : Y 2 is equal to 0 and then we're going
17:47 : to need a y3 at the end to return on the
17:52 : function what is y 3 well Y 3 is going
17:55 : to be what the Phi of Y 1 and y 3 how do
18:00 : I write Phi
18:05 : there's why three is gonna be the fire
18:09 : of y1 and y2 as in well if it came from
18:13 : this branch it's gonna give you y1 if it
18:16 : came from this branch is gonna give you
18:18 : Y 2 so we can always construct this
18:23 : situation in the simplified environment
18:27 : of or in this simple of simple construct
18:30 : of if expressions in this language but
18:34 : that's not going to be the case if you
18:37 : want to support a an if statement in a
18:41 : conventional language like JavaScript or
18:43 : C because they do not have this property
18:47 : in a C or JavaScript you can basically
18:49 : do you know you can you can do arbitrary
18:56 : assignments within within the within
19:03 : either block of the if statement and if
19:10 : you had that stuff you you're gonna need
19:12 : a v function for W you're gonna need a
19:16 : 504 z and so forth so so that problem
19:22 : will be much harder if we had to support
19:26 : that but we don't put a kaleidoscope
19:28 : language and that's the shortcut that we
19:30 : took okay all right I'm gonna stop this
19:33 : episode here in the next episode we're
19:36 : gonna go into the for loop and see how
19:39 : we implement that