cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
JMP is taking Discovery online, April 16 and 18. Register today and join us for interactive sessions featuring popular presentation topics, networking, and discussions with the experts.
Submit your abstract to the call for content for Discovery Summit Americas by April 23. Selected abstracts will be presented at Discovery Summit, Oct. 21- 24.
Craige_Hales
Super User
JSL BLOB in an ESP32 Clock

in What time is it? I started describing a clock-building project. I'm midway (I hope!) through the project, and it is a little different from where it began. Previously the emphasis was on getting the time from the internet, which it might still do as a backup if GPS doesn't get a signal, but the primary time source will probably be a GPS.

But, in either case, time arrives as UTC, not local time. How to convert UTC to local is not a problem on a computer running a real operating system. When a real OS is installed, a timezone is supplied, like America/New_York, etc. That's all the OS needs (along with a table of data about names, offsets from UTC, and when daylight time starts and ends, which is built in to Windows, Mac, and Linux OSs.)

The ESP32 based clock only has about 4MB memory total, and runs a version of FreeRTOS , a real-time OS for tiny computers. It does not include a table to convert from a name of a timezone to the rule for timezone, but it does understand the rules. A rule is a short string that indicates the offset from UTC, whether daylight (summer) time is observed, and when daylight time starts and stops. 

A clock with a GPS should be able to set itself to local time if it has a complete map of lat/lon data to timezone rules. Making that map, and packing it into <2MB is what this JSL does. The comments tell the story of using JSL to build a binary blob of run-length encoded data to flash onto a ESP32 SoC (System On a Chip.) The if(0) sections are needed the first time the code runs; some of them take a long time and did not need to be rerun as the code was developed. Various sections of the JSL needs to run about three times to reach the best-compressed state because the order of the zone drawing changes, which changes the usage count of the run codes, which affects how pixels are obscured in the graph because the least used are drawn last to make sure they show up. 

 

Also: if you use this, the resolution is worse than 1 mile in places. Do your own testing!

 

/*
Problem one: A GPS device returns both the lat,lon location and the 
UTC time. How can you get the local time from that information?

You can't. You also need to know what timezone boundary the lat,lon
falls in. There's a nice github project, updated frequently, with
changes. 
https://github.com/evansiroky/timezone-boundary-builder

There's also one or more web sites like 
https://timezonedb.com that provide an API to some database similar
to the boundaries.

Problem two: Build a clock using an ESP32 and a GPS. How can it get the
timezone? It only has about 2MB of memory and it can only access the
internet if it has wifi access and a password.

Since a clock should "just work", the internet can be the primary
answer, used if available. A fallback answer is needed too; the data 
for the timezones needs to be compressed into a structure that is 
small, sufficiently detailed, and reasonably fast. (I probably will
not use the website because I'd have to bake in my API-key...)

Timezones are identified by IANA with names like America/New_York
(no longer Eastern Standard Time, EDT, DST for daylight saving time, etc
because "lots of places have an east.") America/New_York implies
when the time changes happen, but it requires another table to
get those rules. The names change, sometimes, as do the rules.
America/Godthab changed its name to America/Nuuk in 2020, and
everyone has experienced their local leaders playing with "summer time".
https://data.iana.org/time-zones/tzdb/NEWS

Linux systems have another way to represent timezones, using a rule 
rather than a name. (Going out on a limb here...) a Linux system has
tables for converting names to rules. A tiny ESP32 system only
understands the rules. A rule looks something like
<-03>3<-02>,M3.5.0/-2,M10.5.0/-1 (for America/Nuuk) 
or
EST5EDT,M3.2.0,M11.1.0 (for America/New_York)
I think those are called "posix" rules. EST5 means 5 hours behind
UTC, the EDT means daylight time is observed in the summer. The two
time-like values says when to start and end daylight time using, for
the New York example, third month, second week, sunday with a 2AM default.
Nuuk is 3 or 2 hours behind and uses the last(5) week and 1AM (or
is that 11PM?) Fortunately, we don't need to read or write the rules, the OS does.

The name-based America/New_York and the rule-based EST5EDT,M3.2.0,M11.1.0
provide different benefits for different use cases. The name-based system
uses a database to look up the name and can provide historical information
about whether daylight time was in effect in May of 1970, for example.
The rule-based system needs no database and encodes everything needed
to convert any time the rule covers (maybe not 1970!) between UTC and local.

In this project, I realized a bit late that I needed the rule-based answer;
had the name-based answers been converted to rule-based earlier, collapsing
adjacent names with the same rule into one area instead of several, the
file would be a bit smaller. But, it is small enough.

This is the JSL to build the timezone data structure that is stored
into the ESP32 flash memory (4MB which also holds the C code.) The
C language code on the ESP32 will use the resulting binary data.
*/

// data from https://github.com/evansiroky/timezone-boundary-builder
// https://github.com/evansiroky/timezone-boundary-builder/releases/download/2020d/timezones.geojson.zip

// https://github.com/nayarsystems/posix_tz_db/blob/master/zones.csv to convert America/New_York to EST5EDT,M3.2.0,M11.1.0
// https://github.com/nayarsystems/posix_tz_db/blob/master/zones.csv
// made my own using the gen-tz.py script, modified with freqCodeToName names added, America/Nuuk was missing
/* here's the import of the file that converts from America/New_York to EST5EDT,M3.2.0,M11.1.0
   I open the JMP table with the from and to column names below...

dtTrans = open("f:/ClockScripts/gen-tz.csv",Import Settings(
		End Of Line( CRLF, CR, LF ),
		End Of Field( Comma, CSV( 1 ) ),
		Strip Quotes( 1 ),
		Use Apostrophe as Quotation Mark( 0 ),
		Use Regional Settings( 0 ),
		Scan Whole File( 1 ),
		Treat empty columns as numeric( 0 ),
		CompressNumericColumns( 0 ),
		CompressCharacterColumns( 0 ),
		CompressAllowListCheck( 0 ),
		Labels( 0 ),
		Column Names Start( 1 ),
		Data Starts( 1 ),
		Lines To Read( "All" ),
		Year Rule( "20xx" )
	));*/

dtTrans = open("$desktop/timezoneNameConversion.jmp"); // cols from and to, used way below to make an associative array
/* this file was missing Nuuk, above was the regenerated file including Nuuk (run Python on Linux system on the gen-tz)
open("https://raw.githubusercontent.com/nayarsystems/posix_tz_db/master/zones.csv",Import Settings(
		End Of Line( CRLF, CR, LF ),
		End Of Field( Comma, CSV( 1 ) ),
		Strip Quotes( 1 ),
		Use Apostrophe as Quotation Mark( 0 ),
		Use Regional Settings( 0 ),
		Scan Whole File( 1 ),
		Treat empty columns as numeric( 0 ),
		CompressNumericColumns( 0 ),
		CompressCharacterColumns( 0 ),
		CompressAllowListCheck( 0 ),
		Labels( 0 ),
		Column Names Start( 1 ),
		Data Starts( 1 ),
		Lines To Read( "All" ),
		Year Rule( "20xx" )
	));
*/

if( 0, // about 2 minutes...run once... download from https://github.com/evansiroky/timezone-boundary-builder/releases/download/2020d/timezones.geojson.zip
	za = Open( "z:/timezones.geojson.zip", "zip" );
	blob = za << read( "combined.json", Format( blob ) );
	p = Parse JSON( Blob To Char( blob ) ); // this grinds for a bit, it is 135MB
	Save Text File( "$desktop/rlejson.jsl", char(p) ); // this is what is needed next time
);
/* run the above to generate this file the first time */
P = Eval( Parse( Load Text File( "$desktop/rlejson.jsl" ) ) ); // loads faster this way
/*
this is the only interesting key, grab it once. probably this is what should have been saved/reloaded. 
WATER: there is another version of the geojson that includes oceans. I did not look at it. The
JSL presets the bitmap to 0, then fills in polygons with values > 0 for timezones. This works out 
well; the C array is indexed from 0 and has a entry for water there. The JSL test code detects 
the 0 to avoid an index error and reports "water";
*/
featureList = P["features"]; // list of 426 AAs (no water)
/*
the "P" is for Parsed json. This is the best way to load this .json
data; using the json wizard almost works but does not handle the nested
arrays used to represent disjoint polygons.

The next two files are saved from a prev run. find "if(0, // regenerate" below
to generate them; takes a while and this saves time if nothing else changed.
If you are lucky enough to only run this code once, you won't care. Saved a lot
of development time further down in the file.
*/
rle = Eval( Parse( Load Text File( "$desktop/rletext.jsl" ) ) );
/*
you won't have this file the first time; I think freqCodeToName={"water"}; is the correct choice,
BUT you'll need to do a second run to get good compression! This list *should* have the
"America/New_York" names listed in order of how many times they create a run in the 
run length encoding. The 128 most frequent ones get represented as 1-byte codes, the
remaining 300 or so get a 2-byte code. They get counted and saved in this file later.
*/
freqCodeToName = Eval( Parse( Load Text File( "$desktop/rlename.jsl" ) ) );
/*
Mostly this JSL uses the "America/New_York" names, in the order of the feature list.
This simple-minded list of those names can be indexed by the feature index (usually) or
using loc() (rarely) to get the index from the name.
The for loop ignores the geometry for now and make the iFeatureToName list.
*/
iFeatureToName = {};
for( global:iFeature = 1, global:iFeature <= N Items( featureList ), global:iFeature += 1,
	tzdata = featureList[global:iFeature]; // aa {"geometry", "properties", "type"}
	tzproperties = tzdata["properties"]; // aa ["tzid" => "Africa/Abidjan"]
	tzname=tzproperties["tzid"];
	insertinto(iFeatureToName,tzname);
	if(!contains(freqCodeToName,tzname),
		write("\!nadd ",tzname);
		insertinto(freqCodeToName, tzname); // see comment above about "freqCodeToName={"water"};" -- this makes the uncompressed version
	);
);

// sanity check, will need work or removal in the future.
if( freqCodeToName[1] != "water" | freqCodeToName[2] != "Asia/Shanghai" | freqCodeToName[426] != "Europe/Busingen" | freqCodeToName[427] != "Europe/Vatican",
	Throw( "freqCodeToName" ) // probably comment this out on the first run which won't be sorted, and maybe later if the world changes.
);

/*
two helper functions for the graphics script to call
dopoly does one disjoint part of a timezone.
Further down, the graphbox will be recreated over and over for each global:iFeature because
there is an anti-aliased edge that has pixels of a different color than the center pixels
but they are all the same timezone...so doing the timezones one-at-a-time is the easiest way,
though it will take a long time because the bitmap is 16K x 30K huge.
*/
dopoly = function( {tzName, tzcoords},
	{i, m, xs,ys},
	for( i = 1, i <= N Items( tzcoords ), i += 1, 
		/*
		I'm still not sure what the 2,3,... polys represent. Holes maybe, 
		in a single disjoint poly. Render same color, smaller polys render later (2nd run).
		
		this is part of the data that is not right with the json wizard (comments at top)
		*/
		//if(i>1,fillcolor("magenta")); 
		m = Matrix( tzcoords[i] );
		xs = m[0, 1];
		ys = m[0, 2];
		//pencolor("white");pensize(1);
		polygon/*line*/(xs, ys);// xmatrix,ymatrix
	)
);

/*
drawpolys does all the disjoint parts of one timezone; a timezone with only 1 part is not nested
quite as deep as a timezone with multiple disjoint parts.
*/
drawpolys = function( {}, // 
	{tzdata, tzgeometry, tzproperties, tztype, tzGeometryType, tzcoordinates, code, greenbits, bluebits, redbits, ipoly},
	tzdata = featureList[global:iFeature]; // aa {"geometry", "properties", "type"}
	tzgeometry = tzdata["geometry"]; // aa {"coordinates", "type"}
	tzproperties = tzdata["properties"]; // aa ["tzid" => "Africa/Abidjan"]
	/* sanity checking, these should match the log line that follows from the driver loop below */
	write("\!n drawing ",global:iFeature,": ", tzproperties["tzid"] );
	tztype = tzdata["type"]; // string "Feature"
	tzGeometryType = tzgeometry["type"];//"Polygon" or "MultiPolygon"
	tzcoordinates = tzgeometry["coordinates"]; // list
	Fill Color( "white" );
	if(
		tzGeometryType == "MultiPolygon", /* a disjoint time zone */
			for( ipoly = 1, ipoly <= N Items( tzcoordinates ), ipoly += 1,
				dopoly( tzproperties["tzid"], tzcoordinates[ipoly] ) /* deeper nest */
			);//
	, /*else if*/tzGeometryType == "Polygon", /* single poly timezone */
			dopoly( tzproperties["tzid"], tzcoordinates );// no extra nest
	, /*else*/
		Throw( "tzGeometryType" || Char( tzGeometryType ) )
	);
);
/*
this needs to run the first time to generate the bitmap to get the initial frequencies of the
run length codes. It needs to run the 2nd time too, using the frequency ordering. Maybe.
Originally 1-pixel zones like Europe/Vatican reported their name. Now they report thier
description, and in this case, the surrounding Europe/Rome has the same description. There
could be some other edge cases that still matter.
"composition" is a large matrix (3.5GB), if rerunning this code, set it to 0 to release the
old one before making a new one. otherwise 2*3.5GB must be in memory at the same time.
*/
composition = 0;
composition = J( 16000, 30000, 0 );42;
if(0, // let this code run to rebuild the composition
	for( ifreq = 2/* 1 is water */, ifreq <= N Items( freqCodeToName ), ifreq += 1,
		/*
		this is the big-to-small list of timezones; get the name of the zone to do next...
		*/
		thisname=freqCodeToName[ifreq];
		/*
		this global variable tells the poly routine what to work on; locate the name and use the index from loc
		*/
		global:iFeature = loc(iFeatureToName,thisname)[1];// draw smallest last
		/*
		Surprisingly nothing happens when the graphbox is made without a window, until...
		*/
		gb = Graph Box(
			X Scale( -180, 180 ),
			Y Scale( -90, 90 ),
			framesize( 30000 + 1, 16000 + 1 ), // drawing beyond 30,000 may not always work, 16x30 is about as big as it can be.
			<<backgroundcolor( "black" ),
			drawpolys();
			//marker(colorstate("green"),{-180,-90},{180,-90},{-180,90},{180,90},{0,0}); // corner markers might help figure out trimming, below
		);
		/*
		...the picture/bitmap is requested. That takes some time.
		*/
		pixels = gb[framebox( 1 )] << getpicture; // 30 seconds to make the picture
		gb = 0;
		pixels = pixels << getpixels; // 10 seconds to convert to pixel matrix
		/*
		trimming the image. added 1 in the graphbox to get the data inside a 30000x16000 
		(not 29999x15999) space, now strip off the extra frame pixels
		*/
		pixels = pixels[2 :: (N Rows( pixels ) - 5), 2 :: (N Cols( pixels ) - 5)];
		/*
		after another small sanity check, find the nonzero (rendered white timezone on black background)
		pixels and add them to the composition as *** global:iFeature *** value.
		iFeature will become the RLE code in the next step.
		*/
		if( N Rows( composition ) == N Rows( pixels ) & N Cols( composition ) == N Cols( pixels ),
			composition[Loc( pixels )] = global:iFeature; // 
		, // else
			Throw( "composition" )
		);
		write("\!n freq=", ifreq," composition code=",global:iFeature," for ",thisname );
		if(thisname == "America/New_York", write(" ***"));
		Wait( .1 );
	);
	/*
	delete the old file saving the composition, then recreate it.
	My computer was unable to do it in one save, so I split it 
	into four appends. Probably no real point in the delete since
	the first save is a replace.
	And this hard codes the 16000 height, but that isn't the dimension
	I wish was bigger; the horizontal resolution is more important for
	time zones, most of the time.
	*/
	try(deletefile("$desktop/rlecomp.jsl"));

	bcomp = matrixtoblob(composition[1::4000,0],"float",8,"little");
	file=Save Text File( "$desktop/rlecomp.jsl", bcomp, mode("replace") );
	show(filesize(file));
	bcomp = matrixtoblob(composition[4001::8000,0],"float",8,"little");
	file=Save Text File( "$desktop/rlecomp.jsl", bcomp, mode("append") );
	show(filesize(file));
	bcomp = matrixtoblob(composition[8001::12000,0],"float",8,"little");
	file=Save Text File( "$desktop/rlecomp.jsl", bcomp, mode("append") );
	show(filesize(file));
	bcomp = matrixtoblob(composition[12001::16000,0],"float",8,"little");
	file=Save Text File( "$desktop/rlecomp.jsl", bcomp, mode("append") );
	show(filesize(file));
);
/*
reload the composition created above
*/
composition = 0;
composition = J( 16000, 30000, 0 );
composition[1::4000,0] = blobtomatrix(loadtextfile("$desktop/rlecomp.jsl",blob(readOffsetFromBegin(0),readlength(960000000))),"float",8,"little",30000);
composition[4001::8000,0] = blobtomatrix(loadtextfile("$desktop/rlecomp.jsl",blob(readOffsetFromBegin(1*960000000),readlength(960000000))),"float",8,"little",30000);
composition[8001::12000,0] = blobtomatrix(loadtextfile("$desktop/rlecomp.jsl",blob(readOffsetFromBegin(2*960000000),readlength(960000000))),"float",8,"little",30000);
composition[12001::16000,0] = blobtomatrix(loadtextfile("$desktop/rlecomp.jsl",blob(readOffsetFromBegin(3*960000000),readlength(960000000))),"float",8,"little",30000);
/*
view it for another sanity check
*/
img = 0;
img = New Image( Heat Color( composition / Max( composition ), "spectral" ) );
img << scale( 1 / 16 ); // make it fit screen better. maybe use 1/8 on a 4K display.
New Window( "x", img );

/*
This JSL mostly builds the RLE (run length encoded data), but as a side effect
counts how common the RLE codes are. 
This loop builds uncompressed pairs {code,length} in a list of Nx2-element matrices
rle is a 16000 element list; each element is a run length encoded row of the picture.
each row of rle is a Nx2 matrix of the rle encoded changes for the row.
*/
if(0, // regenerate
	codecounts = [=>0];
	rle = {};
	for( irow = 1, irow <= N Rows( composition ), irow += 1,
		if(mod(irow,100)==0,write(irow," ");wait(.01));
		row = {}; // a list for one row
		code = -1; // initial no match
		length = 0; // initial no length
		for( icol = 1, icol <= N Cols( composition ), icol += 1,
			if( composition[irow, icol] != code, // new run begins
				if( length > 0, // initial no match has length 0
					run = {}; // output code&length of prev run
					codecounts[code]+=1;
					run[1] = code;
					run[2] = length;
					row[N Items( row ) + 1] = run;
				);
				length = 1;
				code = composition[irow, icol];//
			, // else
				length += 1
			)
		);
		run = {};// handle the last one on the row
		codecounts[code]+=1;
		run[1] = code;
		run[2] = length;
		row[N Items( row ) + 1] = run;
		rle[N Items( rle ) + 1] = Matrix( row );// convert to Nx2 matrix and store at end of rle list
	);

	countkeys = codecounts<<getkeys; // 0..426. these are iFeature values.
	countvals = codecounts<<getvalues; // {108243, 849, 653, 1253, ... }
	rnk = rank(countvals);
	countvals = reverse(countvals[rnk]);
	countkeys = reverse(countkeys[rnk]); // these are iFeatures in big to small order
	if(countkeys[1]!=0,throw("countkeys"));
	freqCodeToName = iFeatureToName[countkeys[2::nitems(countkeys)]];
	insertinto(freqCodeToName,"water",1); // 427
	/*
	save for another run if nothing above changed
	*/
	Save Text File( "$desktop/rletext.jsl", Char( rle ) );
	Save Text File( "$desktop/rlename.jsl", char(freqCodeToName) );
);
/*
really close to building the binary data now. this converts the rle list of matrices
into a matrix directory of offsets into a blob, and the blob of compressed code,length
pairs. The compression for code and length is the same for both, 1 or 2 bytes depending
on size of number. partBlob speeds the build up by reducing the number of small appends
to a really huge blob.
*/
RleTableIndex = J( N Items( rle ), 1, . );
RleTableDataBlob = Char To Blob( "" );
address = 0; // increments according to compression and writes to RleTableIndex
irun = 0;
partBlob = Char To Blob( "" );
for( i = 1, i <= N Items( rle ), i += 1,
	if(mod(i,100)==0,write(i," ");wait(0););
	RleTableIndex[i] = address;
	k = 0;
	for( j = 1, j <= N Items( rle[i] ) / 2, j += 1,
		irun += 1;
		k += 1;
		code = rle[i][k];
		k += 1;
		length = rle[i][k];
		if(!(0<=code<=426),throw("code"||char(code)));
		if(!(1<=length<=30000),throw("length"||char(length)));
		partBlob = partBlob || (if( code <= 127,
			address += 1;
			Matrix To Blob( Matrix( code ), "int", 1, "big" ); // big: sign first
		,
			address += 2;
			Matrix To Blob( Matrix( -(code) ), "int", 2, "big" );
		) || if( length <= 127,
			address += 1;
			Matrix To Blob( Matrix( length ), "int", 1, "big" );
		,
			address += 2;
			Matrix To Blob( Matrix( -length ), "int", 2, "big" );
		));
	);
	if(mod(i,100)==0,RleTableDataBlob = RleTableDataBlob || partBlob;	partBlob = Char To Blob( "" ););
);
show(length(RleTableDataBlob));// 1,385,104 // 1,519,228 

/*
build the string table. the binary data needs the names or descriptions that go with
the codes in the rle. There will be a matrix of offsets into a blob of packed strings.
*/
StringTableIndex = J( Nitems( iFeatureToName )+1, 1, . );
null = Hex To Blob( "00" );
StringTableDataBlob = null; // water. represented by a zero-length null terminated string.
StringTableIndex[1] = 0; // pointer to water
/*
at the very top, a table to convert from name to rule was loaded. make a lookup AA
and then translate the feature name to the rule and put the rule in the blob.
You won't care about the Nuuk/Godthab issue, I hope. It shows up here if Nuuk
is unknown by the gen-tz.py code mentioned earlier as key not found. A kludge
fix is shown, it might need more. My fix involved putting all the items in
iFeatureToName into gen-tz.py; that allowed it to lookup the rule in the linux system.
*/
xlate=associativearray(dtTrans:from<<getvalues,dtTrans:to<<getvalues);
// 'America/Nuuk' is 'America/Godthab'
//if(!xlate<<contains("America/Nuuk"),
//xlate["America/Nuuk"] = xlate["America/Godthab"];
//);
for( i = 1, i <= Nitems( iFeatureToName ), i += 1,
	StringTableIndex[i+1] = Length( StringTableDataBlob );
	tznamet=xlate[iFeatureToName[i]];
	StringTableDataBlob = StringTableDataBlob || (chartoblob(tznamet) || null);
	//show(i,chartoblob(iFeatureToName[i]),length(StringTableDataBlob));
);
Length( StringTableDataBlob );//7022
loc(iFeatureToName,"America/New_York");

/*
Almost got it all. Need a little header for the blob that describes the rest of the blob.
the initial ascii-word-picture below is wrong, n is bigger and the offsets are wrong. Read the code instead.
It took a while to understand how to map the flash memory holding this blob into
the ESP32 address space. It shows up at a huge address. This blob needs to use offsets
and not try to use pointers, there is no way to relocate them...not 16,000+ of them.
*/
//////////////// output BLOB /////////////////
// HEADER
// 00000000:        10 4-byte number of table ptrs (string, rle)
// 00000004:        0C 4-byte ptr to string table
// 00000008:  nnnnnnnn 4-byte ptr to rle table (need size of StringTableIndex and StringTableDataBlob)
// STRING TABLE
// 0000000C:   C+4*427 StringTableIndex 427 4-byte ptrs to string table entries, 4 bytes each
// 00000010:       ... 2nd string 4-byte ptr
//  C+4*427:  "a\0b\0" StringTableDataBlob of null term strings
// RLE TABLE
// nnnnnnnn:  rrrrrrrr RleTableIndex 4-byte ptr to start of first run
//      ...:       ... 16,000 4-byte run ptrs 
// n+64,000:  rle      RleTableDataBlob (1-2 compression from above)

stiLength = 4*nitems(StringTableIndex);// 4-byte offsets, string table's length
rtiLength = 4*nitems(RleTableIndex); // 4-byte offsets, run table's length
stdLength = length(StringTableDataBlob);
n=12;// number of 4-byte integers, not counting the first n value, counting the "END." value
// first build a matrix of 4-byte integers. some are offsets, some are values.
// 4+n*4 is the header length, including the initial n value.
header = n // number of offsets and ints that follow:
	||((4+n*4)+0) // offset to string index table
	||((4+n*4)+stiLength) // offset to string data "America/New_York\0" and similar packed strings
	||((4+n*4)+stiLength+stdLength) // offset to rle index table
	||((4+n*4)+stiLength+stdLength+rtiLength) // offset to rle data 1 or 2 compressed
	||((4+n*4)+stiLength+stdLength+rtiLength+length(RleTableDataBlob)) // date. sanity checking. not used for much.
	||nrows(composition) // lat
	||ncols(composition) // lon
	||-90||90 // lat coverage for nrows
	||-180||180 // lon cover for ncols
	||hextonumber(hex(reverse("END.")));// little endian reverses again (hextonumber data is big endian)
if(n!=nitems(header)-1,throw("header"));
/*
convert the header matrix to a blob
*/
headerBlob = matrixtoblob(header,"int",4,"little");
hbLength = length(headerBlob);

/*
Convert the string table index to a blob. NOTE the multiply by 0! these are not
pointers and not relocated offsets. they are relative offsets within the string table data
(distinguish between string table INDEX and string table DATA; the first item in the
INDEX is the offset of the first string in the DATA.)
*/
stiBlob = matrixToBlob(StringTableIndex+0*(hbLength+stiLength/*relocate ptrs*/),"int",4,"little");// NOT relocated, index offsets relative to StringTableDataBlob
if(length(stiBlob)!=stiLength,throw("stiBlob"));

/* ditto the RLE table INDEX and DATA */
rtiBlob = matrixtoblob(RleTableIndex+0*(hbLength+stiLength+rtiLength/*relocate ptrs*/),"int",4,"little");// NOT relocated,... RleTableDataBlob
if(length(rtiBlob)!=rtiLength,throw("rtiBlob"));

/* put it together, in order */
finalblob = headerBlob 
			|| stiBlob 
			|| StringTableDataBlob 
			|| rtiBlob 
			|| RleTableDataBlob 
			|| chartoblob("TZ lookup built "||char(asdate(today()))) || null;

write("\!nfinalblob size=",length(finalblob));// 1592045

// a little testing, not much 
write(
	"\!nstring NY=",
	blobpeek(finalblob,
		blobtomatrix( 
			Blob Peek( finalblob, 
				Blob To Matrix( Blob Peek( finalblob, 1/*string index*/ * 4, 4 ), "int", 4, "little" )[1]/*offset to index*/ + 4 * 152/*offset in index to offset of NY in string data*/
				, 4 
			),
			"int",4,"little"
		)
		+
		Blob To Matrix( Blob Peek( finalblob, 2/*data*/ * 4, 4 ), "int", 4, "little" )[1]/*offset to start of string data*/
		,
		17
	)
);

// a little testing, not much 
write("\!nsignature=",
	blobpeek(
		finalblob,
		Blob To Matrix( Blob Peek( finalblob, 5 * 4, 4 ), "int", 4, "little" )[1],
		300
	)
);

//
//************ save the binary data to be flashed later on the ESP32 *************
//
savetextfile("f:/ClockScripts/tzblob.bin", finalblob);

/*
this code is translated to C on the ESP32, this is a more friendly environment for prototyping...
*/

// quick test
lookup = function({lat=41.91,lon},{row,col,r,c,code},
	row=floor(interpolate(lat,-90,N Rows( composition ),90,1+0))-0;
	r=RleTableIndex[row];
	col=round(interpolate(lon,-180,1+0,180,N Cols( composition )))-0;
	c=0;
	while(c<col,
		code = blobtomatrix(blobpeek(RleTableDataBlob,r,1),"int",1,"big")[1];
		if(code<0,
			code = -blobtomatrix(blobpeek(RleTableDataBlob,r,2),"int",2,"big")[1];
			r+=2;//
		,//
			r+=1;
		);
		length = blobtomatrix(blobpeek(RleTableDataBlob,r,1),"int",1,"big")[1];
		if(length<0,
			length = -blobtomatrix(blobpeek(RleTableDataBlob,r,2),"int",2,"big")[1];
			r+=2;//
		,//
			r+=1;
		);
		c += length;
	);
	if(code>0,
		iFeatureToName[code];
	,//
		"water"
	)
);

/*
draw a bunch of markers, color coded for timezone, over a map.
This give a view of the accuracy near rivers and obvious political boundaries.
*/
//////////////////////////////////////////////////////////////////
markersOver = function({clat,clon,name,grid=1},
	dtVat=New Table( name,
		New Column( "tz", Character, "Nominal" ),
		New Column( "lat", Numeric, "Continuous", Format( "Best", 12 ) ),
		New Column( "lon", Numeric, "Continuous", Format( "Best", 12 ) )
	);

	for(ylat=clat-.5*grid, ylat<=clat+.5*grid,ylat+=.004*grid,
		for(xlon=clon-.5*grid,xlon<=clon+.5*grid,xlon+=.006*grid,
			dtVat<<addrows(1);
			dtVat:tz=lookup(ylat,xlon);
			dtVat:lat=ylat;
			dtVat:lon=xlon;
		)
	);
	dtVat:tz<<label(1);
	dtVat<< Color or Mark by Column(dtVat:tz);
	dtVat<<setdirty(0);
	dtVat = Graph Builder(
		Size( 1843, 976 ),
		Show Control Panel( 0 ),
		Variables( X( :lon ), Y( :lat ) ),
		Elements( Points( X, Y, Legend( 3 ) ) ),
		SendToReport(
//			Dispatch(
//				{},
//				"lon",
//				ScaleBox,
//				{Min( 12.4031559352746 ), Max( 12.5218440647255 ), Inc( 0.01 ),
//				Minor Ticks( 1 )}
//			),
//			Dispatch(
//				{},
//				"lat",
//				ScaleBox,
//				{Scale( "Geodesic" ), Format( "Best", 12 ), Min( 41.8801686871589 ),
//				Max( 41.9397483777795 ), Inc( 0.01 ), Minor Ticks( 4 )}
//			),
			Dispatch(
				{},
				"Graph Builder",
				FrameBox,
				//{Background Map( Images( "Street Map Service" ) ), Marker Size( 6 )}
				{Background Map(
					Images(
						"Web Map Service", "https://ows.terrestris.de/osm-gray/service?",
						"OSM-WMS"
					)
				), Marker Size( 1 ), Marker Drawing Mode( "Normal" )}
			)
		)
	);
);
/*
the smallest timezone *does* align part way under the data. The pixels in the data are bigger than 
this timezone and it falls on the edge of a pixel.
Kansas is interesting because of its western counties.
*/
markersover(41.905,12.445,"vatican city");
///////////////////////////////////////

markersover(38.76012940186213, -85.07358907248562, "America/Indiana/Vevay") ;
markersover(38, -85, "Kentucky",20) ;

markersover(4, -51.5,"French Guiana/Brazil");

markersover(26.528054926367076, 88.5814945236749,"Bangladesh");

markersover(19.413200799245335, 166.6283736471554,"wake island");

markersover(18.925476569622884, -71.7749456680967, "Haiti");

markersover(-54.7, -68.79215603902576, "Tierra del Fuego");

markersover(81.3, 62.4,"Rudolf Island");

markersover(65.84924591016264, -169,"alaska");

markersover(6.297049492023935, 1.7798964984249592,"origin");

markersover(35, -78,"home",100);
markersover(38.5, -101.5,"kansas",5);

Does Indiana really have eleven time zones?  

Part of the testing to see if the blob works to convert lat/lon to a timezone involved making some maps. The JSL for this map makes a grid of data points over a region. Indiana is a state in the middle of the United States; it is on the edge of the change from Eastern to Central time. The wikipedia article (interesting read) suggests how daylight saving time is a pain, far from the center of a time zone, causing counties within the state to make different choices over many years. The zone names may have had subtle differences in the past with when or if daylight time started and stopped, and not-so-subtle choices between Eastern and Central time. These zone names (from column) appear in Indiana:

New_York, Chicago, and Kentucky zones extend way beyond Indiana.New_York, Chicago, and Kentucky zones extend way beyond Indiana.

There are only two distinct rules (the to column) for the time in Indiana. The parts of Indiana that use Central time are close to large cities outside of Indiana that use Central time.

Eleven zones in one state.Eleven zones in one state.

Kansas time zones

Kansas is 15 degrees (a whole timezone) west of Indiana and has a similar issue on it's western border. Four counties choose to use Denver (Mountain) time rather than Chicago (Central) time. This makes a good test with the county outlines to see not only the left-right alignment but the top-bottom alignment as well. State outline shown here. Wikipedia's Kansas story is not as interesting as Indiana (good!)

The blue Denver time is used by four counties on the western side of Kansas. The Red Chicago time is used by most of the state.The blue Denver time is used by four counties on the western side of Kansas. The Red Chicago time is used by most of the state.

Flashing the data to the ESP32:

Linux shell window. I should probably rename my computer. I dropped that subscription years ago after I got to the end.Linux shell window. I should probably rename my computer. I dropped that subscription years ago after I got to the end.

 

 

Finally, a tip-of-the-hat to IANA for their part in making the internet work.

 

Last Modified: Jan 15, 2021 10:50 AM