Replace int-based access for edges with bytes-based access#2986
Replace int-based access for edges with bytes-based access#2986karussell wants to merge 22 commits into
Conversation
| /** | ||
| * Gets the int value at the given index for the given edgeId | ||
| */ | ||
| int getInt(int edgeId, int edgeRowBytesOffset); | ||
|
|
There was a problem hiding this comment.
maybe just int getInt(int edgeId, int offset)? And in the javadocs it should be 'at the given offset' then.
There was a problem hiding this comment.
offset would be a bit confusing in the context with getBytes? And also IMO it is unclear if the offset is int-based or byte-based?
|
|
||
| import com.graphhopper.util.BitUtil; | ||
|
|
||
| public class EdgeBytesAccessArray implements EdgeBytesAccess { |
There was a problem hiding this comment.
Maybe call it ArrayEdgeBytesAccess ? It's an EdgeByteAccess that uses an array rather than an array of EdgeBytesAccess? Just like there is java.util.ArrayList not java.util.ListArray(?).
| public static final int VERSION_EDGE = 22; | ||
| public static final int VERSION_EDGE = 23; | ||
| // this should be increased whenever the format of the serialized EncodingManager is changed | ||
| public static final int VERSION_EM = 3; |
| // measureRouting(hopper, new QuerySettings("routingCH_alt", count / 100, isCH, isLM). | ||
| // withInstructions().sod().alternative()); | ||
| // measureRouting(hopper, new QuerySettings("routingCH_with_hints", count, isCH, isLM). | ||
| // withInstructions().sod().withPointHints()); | ||
| // measureRouting(hopper, new QuerySettings("routingCH_no_sod", count, isCH, isLM). |
There was a problem hiding this comment.
We shouldn't forget to uncomment this before merge?
| @@ -1,5 +1,7 @@ | |||
| ### 10.0 [not yet released] | |||
|
|
|||
| - replaced IntsRef with BytesRef. Note that setInt(edgeId, index) uses a byte index now, i.e. instead of edgeAccess.setInt(edgeId, index, value) you now have to use edgeAccess.setInt(edgeId, index * 4, value) | |||
There was a problem hiding this comment.
Maybe mention that previous TagParsers will no longer compile, but since they only pass on the edge access this is easy to fix. The direct usages of setInt are rare in comparison and probably hardly affect any users(?).
| @Override | ||
| public double getCost(DecimalEncodedValue encodedValue) { | ||
| intsRef.ints[0] = turnCosts.getInt(turnCostPtr() + TC_FLAGS); | ||
| return encodedValue.getDecimal(false, -1, edgeIntAccess); |
There was a problem hiding this comment.
I thought using -1 made it a bit more clear that we are passing an unused value? In the code above we still use -1
| */ | ||
| @Override | ||
| public int compareTo(IntsRef other) { | ||
| public int compareTo(BytesRef other) { |
There was a problem hiding this comment.
(Why) do we need to implement Comparable here?
This reverts commit ab278cd.
2 participants
Currently we are only able to read integers from
edges, which is a bit limiting as everything has to fit in one integer and if not, it will require two integers (instead of expanding to just 5 bytes).So, with this change we can use only 5 bytes for the georef (see #2978) and we do not have to use int-based edge flags too, i.e. potential less space for this either. I.e. this change can reduce the storage requirement by 6 bytes (e.g. if flags use 5 or 9 bytes).
There is only a minor slowdown for routingCH_full, probably because of the change of the georef to bytes. The other minor slowdown for routingCH was compensated via removing the new instance of EdgeIntAccess for every EdgeIteratorState (independent of this PR).
Later this change will help if other edge properties like the edge IDs or link ID grow beyond 4 bytes.
TODO:
nodestoo, then we could reduce 4 to probably just 1 byte for the turn restriction information