Strange behavior with summary function

Hello,

I’m working on providing some LLDB integration for the Dylan language (http://opendylan.org/).

I’ve tried using both the versions of lldb that ship with the current release of Xcode5 (lldb-310.2.37) as well as the version that ships with the Xcode6 beta (lldb-320.3.100).

For what I’m currently testing, we generate C which is then compiled by clang.

We have

typedef void* dylan_value;

typedef struct _dylan_byte_string_0 {
dylan_value class;
dylan_value size;
char data[_size + 1];
} dylan_byte_string;

The strange behavior is this:

(lldb) print Kaccessor_read_intoXYstreams_internalsVio.debug_name_
(dylan_value) $2 = 0x005477c8 {<error: unable to read data>}
(lldb) print (dylan_value)Kaccessor_read_intoXYstreams_internalsVio.debug_name_
(dylan_value) $3 = 0x005477c8 {: size: 0}
(lldb) print Kaccessor_read_intoXYstreams_internalsVio.debug_name_
(dylan_value) $4 = 0x005477c8 {<error: unable to read data>}
(lldb) print (dylan_value)Kaccessor_read_intoXYstreams_internalsVio.debug_name_
(dylan_value) $5 = 0x005477c8 {<error: unable to read data>}
(lldb) print (dylan_byte_string*)Kaccessor_read_intoXYstreams_internalsVio.debug_name_
(dylan_byte_string *) $6 = 0x005477c8 {: size: 19, data: accessor-read-into!}

Note that there should be no difference between $2, $3, $4 or $5. The value is a dylan_value but notice that in instance $3, it says the size was 0, so it was able to read the data (although incorrectly). When I cast it through to the “dylan_byte_string*”, it read the data correctly.

The Python code for dealing with a string is:

def dylan_byte_string_summary(value, dict):
size = value.GetChildMemberWithName(‘size’).GetValueAsUnsigned() >> 2
if size == 0:
return ‘{: size: 0}’
else:
data = get_byte_string_data(value)
return ‘{: size: %d, data: %s}’ % (size, data)

def get_byte_string_data(value):
target = lldb.debugger.GetSelectedTarget()
byte_string_type = target.FindFirstType(‘dylan_byte_string’).GetPointerType()
value = value.Cast(byte_string_type)
size = value.GetChildMemberWithName(‘size’).GetValueAsUnsigned() >> 2
if size == 0:
return ‘’
data = value.GetChildMemberWithName(‘data’).GetPointeeData(0, size)
error = lldb.SBError()
string = data.GetString(error, 0)
if error.Fail():
return ‘<error: %s>’ % error.GetCString()
else:
return ‘%s’ % string

The relevant configuration is done here:

debugger.HandleCommand(‘type summary add dylan_byte_string -F dylan.dylan_byte_string_summary -w dylan’)
debugger.HandleCommand(‘type summary add dylan_value -F dylan.dylan_value_summary -w dylan’)

When we call dylan_value_summary, we check to see if it is a tagged integer otherwise we dig out the class pointer, get the name of the class and then dispatch to the appropriate summary function … so when we get results like $2, $4 and $5 above, that means we got an error trying to read the class name. I don’t understand how we can get error #3.

In fact, on a subsequent run of the debugger, I just tried printing the exact same thing a couple of times without type casts:

(dylan_value) $0 = 0x005477c8 {: size: 19, data: <error: unable to read data>}
(lldb) print Kaccessor_read_intoXYstreams_internalsVio.debug_name_
(dylan_value) $1 = 0x005477c8 {<error: unable to read data>}
(lldb) print Kaccessor_read_intoXYstreams_internalsVio.debug_name_
(dylan_value) $2 = 0x005477c8 {<error: unable to read data>}
(lldb) print Kaccessor_read_intoXYstreams_internalsVio.debug_name_
(dylan_value) $3 = 0x005477c8 {<error: unable to read data>}
(lldb) print Kaccessor_read_intoXYstreams_internalsVio.debug_name_
(dylan_value) $4 = 0x005477c8 {<error: unable to read data>}
(lldb) print Kaccessor_read_intoXYstreams_internalsVio.debug_name_
(dylan_value) $5 = 0x005477c8 {: size: 19, data: accessor-read-into!}

My favorite one is this one though, as I can’t even see how that’s possible given my code:

(lldb) print Kaccessor_read_intoXYstreams_internalsVio.debug_name_
(dylan_value) $15 = 0x005477c8 {}

Does anyone have any idea about this?

Thanks,

  • Bruce

Of course, I ended up figuring this out. I wasn't adding 1 to the size, so
I wasn't reading the NULL into the SBData, so when calling |GetString| on
the SBData, it was randomly failing. I suspect that perhaps
SBData::GetString can read one past the end of the buffer, so sometimes it
was a NULL and sometimes not?

It would be nice / more usable if the error in this situation indicated
that the data wasn't NULL-terminated rather than a generic "unable to read
data".

At any rate, with this fixed, things look much better. :slight_smile:

- Bruce

I am glad you figured this out!

Feel free to file a bug and give us a quick struct in a C file that can reproduce this issue. We can take a look at fixing any weird SBData behaviors.

Greg