Posts by lenidcamper

    sorry I did not scroll down in the include.


    Based on the log, the string compare (without leading zero) should have exported 2 different models.


    But it is actually exporting <[string8][geometry][..\objects\Piers\Pier_Wood_1_6\pier_wood_1_6]> ???


    I don't see where that could possibly come from based on the included config.

    "1_6" does not appear anywhere in the config!!!


    As I mentioned in the postscript to a previous post, the expression engine did not internally resolve data types of

    variables and all variables defaulted to string. So it was doing a string compare for all numeric variables in expressions.


    I am testing a fix that loads int or float variables as required.


    In the pier example what is the expression in the model map?

    Hi Thomas


    A number of issues here


    At first I thought


    "Of course, you are not using the operator alias #GE# for >="


    The expression engine provides the alias/escape strings for operators to avoid special characters in XML. However >= will actually evaluate correctly.

    "<=" will give a syntax error. for consistency it is probably better to always use the alias strings


    The (...) construct is intended to define precedence in the expression. The ($pier:width) defeats my variable preload algorithm.

    I am parsing the expression into tokens on spaces to pick out variable names to preload before tag data is loaded. This avoids a "variable not defined" exception from the expression engine and allows the sparse tag negation functionality added in 0.7.1


    Use $pier:width #GE# '10' or ($pier:width #GE# '10')


    That lead me to the core problem. It appears there is a bug in the XSharper engine with comparison of left justified numeric values.


    ($height #LE# '90') will evaluate true for height tag value '100'

    but

    ($height #LE# ' 90') will evaluate to false


    I need to look under the hood of the XSharper engine to see if there is any way to fix this or find a workaround.


    /Stu


    PS


    what tag data are you comparing?


    zero padding will work.


    '090' will evaluate correctly compared with '100'

    where '90' will not.


    If you have added these tags using leading zero pad



    PPS

    turns out XSharper variables were defaulting to string type.

    need to detect tag datatype and force a cast to int or float as the variable is set.

    I have a request for parking density:

    Thomas


    I have looked at the parking density issue. Adding a computed parameter is possible but is a can of worms I would rather not open right now.

    Particularly since there is a work around for this issue.


    All it requires is separate configurations for the few (3? 5??) different parking densities.

    Each configuration would have a selectionCondition expression with the appropriate parking_density value and randomInteger upperLimit.

    Using sharedModelMaps there will only be a dozen lines for each configuration.


    If there is a scenario that cannot be solved without numeric tag operations I can look at the issue again.


    /Stu

    could it already work?

    There is currently no way to specify arithmetic operations on tags. The XSharper expression engine has that capability. I only exposed what was needed for boolean selection expressions.


    I need to think about a generic way to expose this.

    Under Windows \ or / might both work

    I had not placed any attention towards platform independence. I can update examples and documentation to forward slash and lower case.

    For me, the r22 throttle now seems to be connected with the pedals.

    I agree the R22 seems twitchier in non profi mode.

    Especially in the first few seconds.

    I have to start in paused state or jump right on the controls to gain stability.

    This may be more realistic, but certainly more challenging than in previous releases.


    PS

    I do like the pilot body in the pilot camera. :thumbup::thumbup:

    With the alpha release I have encountered very distinct "slow motion events"


    At the point of the vertical dotted line frame rate dropped from a very stable 90 down to 11 with a corresponding increase in frame time and a decrease in GPU usage.


    CPU usage and memory usage were unchanged.


    This happens at somewhat regular intervals so I though it was scenery load issue but there was no discernible disk access during the event


    The event happened for about 10 seconds as seen from the graph. When "normal" rendering resumed I immediately exited AFS2 to see what Afterburner captured.


    During the event, sound and instruments were unchanged. The perceived forward motion dropped significantly. Head movement gave blurred rendering .


    There were brief glitches in previous releases that I attributed to scenery load, but nothing on this scale.


    This is with Vulkan rendering engine in VR with anti-aliasing on.



    PS


    To test further I rolled back to the beta release and flew the same path. The problem occurs in beta as well.

    It happened while crossing a level 10 boundary. I removed my custom elevation data and the problem went away.

    I am not sure what is wrong with the elevation data because there does not seem to be any discontinuity in the elevation mesh. (and is not on the custom elevation boundary)

    More investigation is required.


    So the culprit is custom scenery for some reason and not the alpha release.

    In fact the alpha release dealt with the issue better than the beta.

    In alpha it was like flying through glue for 15 seconds.

    In beta the controls locked up and it would not even respond to controller exit.

    It is possible with the current release of ObectGen to place a building object at the "midpoint" of a building footprint.


    As with the scenProc FLENGTH the wayLength derived tag could be used to distinguish residential vs commercial.


    The random tag could be used to mix model types.


    Currently the objects would have random orientation.

    It is difficult to derive a meaningful orientation for an arbitrary polygon. If the way has 5 nodes it is a square or rectangle and an axis of orientation could be derived. But few houses are simple rectangles (particularly in NA housing developments).

    Uploaded version v0.7 of ObjectGen


    The main feature is XREF library support.

    There are some schema changes to support this.

    There is a standalone xmlConvert utility to migrate locally modified configuration files to the current schema.


    See the readme and User Guide


    I have seen examples of TSC files with enumerated [elements] and with [0]

    I assumed the scenery loader needs to merge [element] entries from multiple TSC files into an internal list anyway so enumeration does not seem

    to be particularly relevant. Since it works without enumeration I did not bother to add it.


    As for the [autoheight_override] I am unclear on the interpretation. I forget where I copied the original snippet for the template but it did have [-1]

    Models I have placed at zero elevation in Sketchup seem to render at grade level with it set to [-1] (or [10] ) so I left it at [-1]


    For an object model with larger area, if there is a slight slope in the elevation data one edge may be floating or depressed.

    I had to extend the pylon piers well below grade to avoid floating pylon legs on steeper grades


    If [autoheight_override] is set to [0] only XREF objects will render (no user defined objects).


    As with the element index the autoheight_override did not seem broken so I did not fix it.


    Changing the default values is easy if there is a "correct" answer.


    Adding an optional model configuration element is more work so I would need to see good justification for the change.

    The models I have converted worked fine with a zero elevation in Sketchup and autoheight_override of -1


    Stu

    can you introduce an empty parking lot factor?

    You don't need dummy map entries.


    If you set the default model name to empty string and set the upperLimit on the random tag to be appropriately larger than the number of (real) entries in the model map, when there is no match in the model map the empty default model is exported.


    In v0.7 the [tmsimulator_scenery_object][element] will no longer be exported if the model name is blank (to avoid the "..\..\..\..\objects\.tmb' not found"messages in the TM.log)

    I had assumed that the default model was fail safe and would always exist , but that is no longer the case.


    Maybe when Jeff is farther along with vehicle animation

    the preview of Aerofly Life is very interesting.


    It may make ObjectGen traffic redundant.
    No indication yet of how the traffic is defined or how easy that will be for large areas.

    The same issues of multiple lanes and intersections need to be addressed.