Using uint64_t for the value of the offset was just wrong. According
to the POSIX standard off_t is a signed integer type with unspecified
size. Here we stick to a 64 bit integer, because this size guaranties
optimal packing of the kshark_entry structure.
Using unsigned values for the timestamps is also a source of problems
and has been a reason for the introduction of multiple bugs in the past.
In principal the value of the timestamps cannot be negative. However,
this value must have the same type as the values used to define the
state of the visualization model, like the range of the model or the
size of the bin. The model state definitions should not take negative
values as well, however their values are recalculated automatically when
the user browses the data and those calculations may result in negative
values in some corner cases. Because of this it is better to use a
signed integer type and treat the negative values as an indicator
of an error rather than have the negative result of the calculations
casted into unsigned type which results into unpredictable behavior of
the model.
Signed-off-by: Yordan Karadzhov (VMware) <y.karadz@xxxxxxxxx>
---
src/libkshark.h | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/src/libkshark.h b/src/libkshark.h
index 0d6c50d1..9eecc2d0 100644
--- a/src/libkshark.h
+++ b/src/libkshark.h
@@ -61,7 +61,7 @@ struct kshark_entry {
int32_t event_id;
/** The offset into the trace file, used to find the record. */
- uint64_t offset;
+ int64_t offset;
/**
* The time of the record in nano seconds. The value is taken from
@@ -69,7 +69,7 @@ struct kshark_entry {
* dependent. The time usually is the timestamp from when the system
* started.
*/
- uint64_t ts;
+ int64_t ts;
};
/** Size of the task's hash table. */
--
2.25.1
|