[play] Add code for mmap in Go and C #32

- need to deal with raw bytes
- can't put c and golang file in one package, go will try to compile the c file

.gitignore

@@ -26,3 +26,4 @@ _testmain.go
 .idea
 vendor
 .vscode
+a.out

doc/survey/lmdb.md

@@ -0,0 +1,45 @@
+# LMDB:
+
+## Ref
+
+- [Howard Chu - LMDB The Databaseology Lectures - CMU Fall 2015](https://youtu.be/tEa5sAh-kVk?t=24m36s)
+
+## Design
+
+- similar to append only B+
+  - the root noode is always written last
+    - this really **likes Akumuli**
+  - but disk space grows without bound
+- WAL    
+- only have two fixed root nodes
+  - pick the one is newest when cold start
+  - [ ] because they just need two versions?
+- two B+ trees per root node
+  - one store user data
+  - one keep track of free space
+- single writer concurrent with many readers
+- transaction handling 
+  - skipped
+  - a memory mapped reader table
+
+Dark Underside
+
+- assume uniform record size
+- can't use sibling pointers
+
+Page Pointers
+
+- naive code mallocs pointer structs to point to pages and their parents
+- LMDB: fixed size stack
+
+Cursor Tracking
+
+- BerkeleyDB sets a tombstone
+- SQLite3 invalidates all open cursors
+- LMDB updates all open cursors
+
+Turn read ahead off?
+
+- shadow paging?
+
+Multi thread, faster than Memached because no lock?

playground/mmap-c/mmapdemo.c

@@ -0,0 +1,48 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <sys/types.h>
+#include <sys/mman.h>
+#include <sys/stat.h>
+#include <errno.h>
+// http://beej.us/guide/bgipc/output/html/multipage/mmap.html
+
+int main(int argc, char *argv[])
+{
+    int fd, offset;
+    char *data;
+    struct stat sbuf;
+
+    if (argc != 2) {
+        fprintf(stderr, "usage: mmapdemo offset\n");
+        exit(1);
+    }
+
+    if ((fd = open("mmapdemo.c", O_RDONLY)) == -1) {
+        perror("open");
+        exit(1);
+    }
+
+    if (stat("mmapdemo.c", &sbuf) == -1) {
+        perror("stat");
+        exit(1);
+    }
+
+    offset = atoi(argv[1]);
+    if (offset < 0 || offset > sbuf.st_size-1) {
+        fprintf(stderr, "mmapdemo: offset must be in the range 0-%d\n", sbuf.st_size-1);
+        exit(1);
+    }
+
+    data = mmap((caddr_t)0, sbuf.st_size, PROT_READ, MAP_SHARED, fd, 0);
+
+    if (data == (caddr_t)(-1)) {
+        perror("mmap");
+        exit(1);
+    }
+
+    printf("byte at offset %d is '%c'\n", offset, data[offset]);
+
+    return 0;
+}

playground/mmap/demo.txt

@@ -0,0 +1,4 @@
+abc
+123
+456
+hahaha

playground/mmap/mmap_test.go

@@ -0,0 +1,43 @@
+package mmap
+
+import (
+	"testing"
+	"os"
+	"syscall"
+)
+
+// TODO:
+// - write
+// - write & read struct
+// - write in golang, read in C/C++
+
+func TestMmap_Read(t *testing.T) {
+	// C source files not allowed when not using cgo or SWIG
+	f, err := os.Open("demo.txt")
+	if err != nil {
+		t.Fatal(err)
+		return
+	}
+	stat, err := f.Stat()
+	size := stat.Size()
+	// NOTE: from https://github.com/google/codesearch/blob/master/index/mmap_linux.go#L19
+	// we can't handle single file larger than 4GB, due to the len function does not work with slice longer than 2^32
+	// TODO: why not use int64(int(size)) == size
+	if int64(int(size+4095)) != size+4095 {
+		t.Fatalf("%s: too large for mmap", f.Name())
+		return
+	}
+	n := int(size)
+	if n == 0 {
+		t.Fatal("file is empty!")
+		return
+	}
+	// TODO: why codesearch is using (n+4095)&^4095? align to 4KB? I think so
+	t.Log((n+4095)&^4095)
+	data, err := syscall.Mmap(int(f.Fd()),0, n, syscall.PROT_READ, syscall.MAP_SHARED)
+	if err != nil {
+		t.Fatalf("mmap %s: %v", f.Name(), err)
+	}
+	t.Log(len(data))
+	t.Log(string(data[0:10]))
+}