0-byte WPTs

[yakra]

0-byte WPT files crash the C++ siteupdate program.
I haven't tested siteupdate.py to see how it copes.

[yakra]

C++
The first datacheck (HIDDEN_TERMINUS) crashes when trying to access the front() of an empty vector.

My first thought is to move the "fewer than 2 waypoints" error check earlier in the process.

Python
siteupdate.py makes it all the way there (ERROR: Route contains fewer than 2 points: ns.ns001 (0 total points)), so nothing to do here.

[yakra]

Take a closer look at what happens to the lines vector in Route::read_wpt.

[yakra]

Paraphrasing for -> while for expository purposes

file.seekg(0, std::ios::end);				//
unsigned long wptdatasize = file.tellg();		// wptdatasize = 0
file.seekg(0, std::ios::beg);				//
char *wptdata = new char[wptdatasize+1];		// 1 byte allocated
		// deleted after processing lines	//
file.read(wptdata, wptdatasize);			// 0 bytes read
wptdata[wptdatasize] = 0; // add null terminator	// wptdata == ""
file.close();						//

// split file into lines				//
std::vector<char*> lines;				//
size_t spn = 0;						//
char* c = wptdata;					//
while (*c)						// *c == 0; skip
{	spn = strcspn(c, "\n\r");			//
	while (c[spn] == '\n' || c[spn] == '\r')	//
	{	c[spn] = 0;				//
		spn++;					//
	}						//
	lines.emplace_back(c);				//
	c += spn;					//
}							//
lines.push_back(wptdata+wptdatasize+1);			// lines has 1 element, pointing to 1 past EOF

// process lines					//
for (unsigned int l = 0; l < lines.size()-1; l++)	// l == lines.size()-1; skip
// ...							//
delete[] wptdata;					// 1 byte deleted

// per-route datachecks					//
if (point_list.size() < 2) el->add_error(		//
 "Route contains fewer than 2 points: " + str());	// this happens
else { /*datachecks*/ }					// this doesn't

std::cout << '.' << std::flush;				// and then the function returns. We're good!

[yakra]

Even after e90d92c, Valgrind reports an invalid read.
In theory, no-harm-no-foul -- The endchar pointer decreases until it points to a non-zero byte. If endchar ends up out of range, we continue, and no writes occur.

It'd still be nice to quiet Valgrind down though; save any future developers from looking at its results & asking "What's this?"
Plus, in the rarest of circumstances, maybe this could cause crashes or unexpected behavior...

Makes sense for efficiency's sake to do this on a per-file basis rather than per-line.

// process lines
for (unsigned int l = lines[1] < wptdata+2; l < lines.size()-1; l++)

looks good at first, but for 0-byte files, there is no lines[1], so we'd need to abort before getting here. No biggie.

We could

file.seekg(0, std::ios::end);
unsigned long wptdatasize = file.tellg();
// CHECK HERE FOR !wptdatasize
file.seekg(0, std::ios::beg);

But what if a file starts with a null zero byte? Odds of this happening are pretty close to nil, but why not...

// split file into lines
std::vector<char*> lines;
size_t spn = 0;
for (char* c = wptdata; *c; c += spn)
{	for (spn = strcspn(c, "\n\r"); c[spn] == '\n' || c[spn] == '\r'; spn++) c[spn] = 0;
	lines.emplace_back(c);
}
// CHECK WHETHER lines.empty()
lines.push_back(wptdata+wptdatasize+1); // add a dummy "past-the-end" element to make lines[l+1]-2 work

There was at one point a .list file with null zeros, so why not add a little more safety.
We just burn off a few more CPU cycles before getting there, but again no biggie. There's only 3 of these errors in HighwayData to worry about ATM.
This would involve a new ErrorList error: File foo.bar.wpt is empty or begins with null zero.
Not going to add this to the Python version, as Route contains fewer than 2 points would still get flagged. These potential errors being so rare, I don't feel a need to have the C++ & Python versions behave exactly the same.

[yakra]

0822708 can cause a double free.
Need to return;