nasa / trick Goto Github PK

Specifically for those in trick_algorithms

Data::Dumper is not always installed by default. We haven't used the module in a very long time. We should remove the dependency.

Need to add something like:

${ALL_SWIG_OBJECTS} : | $(LIB_DIR) $(OBJECT_DIR)

generated Makefile_swig. There is no guarantee that make will not attempt to link in compiled swig file before the standard objects deps are processed.

Monte Carlo dry run file cannot be used to drive monte carlo. It's supposed to be able to do this.

Suppose that in some sim object "foo" you have an array:
double D3x3 [3][3];
and a pointer :
double (*pD3) [3];
and you point to the array like so:
pD3 = D3x3;

Although ref_attributes("foo.pD3[0]") will return the correct address,
ref_attributes("foo.pD3[1]") and ref_attributes("foo.pD3[2]") will not.

The problem is that Trick::MemoryManager::ref_dim() is not correctly calculating the sizes of the elements of an unconstrained array.

Add the scripts that we use to package releases.

Sim will crash if a var_add command is received with a class/structure.

Create rules in the makefile that will compile the er7_utils and fermi-ware directories if they exist. If they do not exist, then don't try and compile them.

Add batch files to bring up GUIs files under Windows.... happy now?

The intent is create a model sim example.
Create a README file that clearly describes the simulation.
Also ensure that the code and the documentation are clear and consistent.

The function "fprintf" does not belong to the list of async-signal-safe functions.
Would you like to clean-up your function "protect_ip_hang"?

I have looked at a few source files for your current software. I have noticed that some checks for return codes are missing.

Would you like to add more error handling for return values from functions like the following?

• ftell ⇒ Csv::peek
• pthread_cond_init ⇒ Threads constructor
• pthread_create ⇒ tc_multiconnect
• sigaction ⇒ MonteCarlo::slave_kill_run
• strdup ⇒ DPC_column_info constructor

I had no problems building Trick on Fedora 21 until I ran sudo make install. The makefile is looking for er7_utils. Recommend either include er7_utils, or remove the dependency from the makefile.

When running a monte carlo set, don't delete the log files in the master run directory.

Trick will not compile out of our new home here on github.

There are ifdefs for DMTCP scattered throughout the code. Consolidate all of these into the DMTCP directory.

I'm using Fedora 21. Java 1.8.0 OpenJDK is in the Fedora 21 repository, not 1.7.0. But either way, java-1.8.0-openjdk did not install the required javac executable. I had to install java-1.8.0-openjdk-devel. Recommend updating the wiki to include supported Linux distributions and versions and confirming that java-1.7.0-openjdk package installs javac or if it requires java-1.7.0-openjdk-devel as well.

In preparation for v15.0.0 release.

We unintentionally have overloaded time and time_0 in the ABM_Integrator class. We can safely remove the overloaded fields.

An extra null pointer check is not needed in functions like the following.

• DPC_delta_curve
• DPC_product
• DPM_column
• DPM_relation
• GPViewProductNode

Can't access fields in REGULA_FALSI structure in python.

I suggest to reuse a higher level build system than your current make file so that powerful checks for software features will become easier.

• CMake
• Autotools

trick_dp should test if fxplot exists or not. If it does not it should desensitize the Fermi plot option and default to Java plotter.

I would like to point out that identifiers like "_ATTRIBUTES_H_" and "_CLOCK_HH_" do not fit to the expected naming convention of the C++ language standard.
Would you like to adjust your selection for unique names?

We have some files that we should exclude from the release like the .git* files and documentation source files that are not pdfs.

Th Digest::MD5 module is not always installed by default on all systems. We only it use it to generate a unique file name. Using the file inode information will serve the same purpose.

make fails with:

cp -r /usr/lib64/llvm/clang/3.4.2 lib/clang
cp: cannot stat `/usr/lib64/llvm/clang/3.4.2': No such file or directory

Centos 6.6
clang/llvm 3.4.2

The source of the error is at:
trick_source/codegen/Interface_Code_Gen/makefile: cp -r $(CLANG_LIBDIR)/clang/$(LLVM_VERSION) lib/clang

Manually changing CLANG_LIBDIR from "/usr/lib64" to "/usr/lib" fixes the problem. The CentOS 6 yum repo puts both the 32 bit and 64 bit libraries at /usr/lib/clang/3.4.2.

The problem is that TRICK_HOST_TYPE and TRICK_HOST_CPU are not defined.
Now need to include ${TRICK_HOME}/makefiles/Makefile.common

This simulation models a parachutist jumping from from a very high altitude, similar to that of Austrian daredevil Felix Baumgartner on Oct 14, 2012.

install_user script does not determine bash, [t]csh correctly.

The message_publish calls that were put in the MemoryManager are too invasive. Replace these calls with stream output

Change log should be in markdown.

If there is a permission issue for writing the makefile, such as the directory has no write permission, no error is given besides "no rule to make target".

If you have this situation.

struct Astruct ;

typedef struct Astruct { } Bstruct ;

then io_src code for Bstruct will not be generated. People want to use the Bstruct name more than the Astruct.

The following conditions will generate a problem with the variable server restarting:
1.) User must specify a port via trick.var_server_set_port(<port_id>) in the input file
2.) Execute simulation and dump a checkpoint on machine A.
3.) ssh to machine B.
4.) Execute simulation and load the checkpoint.

Variable Server will fail with error:
ERROR: Could not establish listen port <port_id> for Variable Server. Aborting.

From this point on, any attempts to connect to the variable server will fail. however, already established connections seem to succeed.

It would be handy if Trick View subclasses could save additional information to .tv files. Let's modify the framework a bit to support that.

Using the ball sim, the sim should run for a bit and take a checkpoint. then continue for a bit, reload the checkpoint and continue to the end. The results should be the same at the end.

By "barf", I mean that many variables available in trick_qp do not show up due to the "->" characters. This is an issue when doing frame logging.

The dry run file is written in ascii where as the values coming out of the random number generators are fed into monte carlo runs as exact binary representation of double numbers. To get these to match, we can change everything to use ascii.

If you have this

class A {
  public:
    int ii ;
} ;

class B : public A {
  public:
    int ii ;
} ;

Only B:ii will show up in the checkpoint and it will be duplicated. We need to support overloaded names.

Before we became open source we have various copyright notices and header comments that are now obsolete. Also remove $Id$ comments as they will no longer be updated with git.

Messages aren't being printed when an event fires. This worked in Trick 10. I do have the following in my input deck:

trick.set_event_info_msg_on()

make install copies the binaries to the install directory, but trick_dp fails to file gte.pm and terminates. gte.pm is found in trick/bin/pm/gte.pm.

Centos 6.6

What is the preferred deployment method? Make install or set the path to the as built bin directory? If preferred deployment is make install, recommend adding additional steps to satisfy runtime binaries. If other method is preferred, recommend either updating the launch scripts or update installation instructions.

Remove the "thread_version" ( from back in the days of Razor) and "service_issues" which hasn't been used forever.

ICG does not compile using clang 3.6

Going open source is a big milestone. Increment the major release number to mark the joyous occasion.

Our release is going to be a tar.gz file now. This will not include a pre-build ICG. So that means everyone needs clang/llvm and we need to ensure the configure script sets that up.