CHANGELOG

The PutForm method send the PUT request with params set in body using content type "application/x-www-form-urlencoded".

The PutFormData method send the PUT request with params set in body using content type "multipart/form-data".

In the next Go release (1.20), parsing URL with invalid percent escape will not throw error anymore [1].

[1] https://github.com/golang/go/issues/56732

In the next Go release, the math/rand is globally seed randomly [1]. This cause our example for Random is always fail.

The fix is to seed it manually using predefined number.

[1] https://github.com/golang/go/issues/54880

This is to fix double slash in case the ServerUrl end with it and the request path begin with it. For example, "http://127.0.0.1/" and "/" should send request to "http://127.0.0.1/" not "http://127.0.0.1//".

If ~/.ssh/config contains a Host section with IdentityFile directive specified and SSH_AUTH_SOCK environment variable is also set, but ssh_config section does not specify "IdentityAgent none" explicitly, the NewClientFromConfig returns an error saying

This changes fix this by dialing remote server twice.

Current implementation exposes dirEntry.filename as fs.DirEntry.Name(). However fs.DirEntry.Info().Name() is always empty string.

In NewClientFromConfig, if Client connect using IdentityFile instead of agent and its success, we add the private key to agent directly.

Unfortunately, since we did not know which key is being negotiated, we add all private keys parsed from IdentityFile.

This is to make the method compatible with ssh.PublicKeysCallback.

Each parsed and unsigned IdentityFile is stored in field PrivateKeys, replacing the Signers field (which is conflict with method names).

This fix tests that sometimes fail when running with -count=X, where X > 1,

Upon inspecting, when client sending larger payload, for example 65536 bytes, server sometimes only read half of them and return an error "resource temporarily unavailable" or "interrupted system call".

The Client have method send that check if the underlying connection (conn) has been closed or not. Since the conn can be closed anytime, for example server send to the control CLOSE frame:

we need to guard the conn with Mutex before calling send to prevent data race.

When calling Close, we initialize the internal channel gracefulClose so the client can check the close response from server on another goroutine serve() and signal back to Close to continue the closing process.

This channel is never closed and nil-ed again after Close which may cause resource leaks.

While at it, use sendClose to minimize duplicate code.

If both exp and got types are string and its longer than 50 chars, it will use the text/diff.Text to show the difference between them. The diff output is as follow,

The "<LINE_NUM> - " print the line number in exp followed by line itself. The "<LINE_NUM> + " print the line number in got followed by line itself. The "^<COL_NUM> - " show the character number in exp line followed by deleted string (or string that not exist in got). The "^<COL_NUM> + " show the character number in got line followed by inserted string (or string that not exist in exp).

Since we only need to call Log and Fatalf during Assert, no need to pass the whole instance of testing.T to Assert. By changing it to Writer, we also can test the Assert.

This remove the AssertBench, because it have the same function parameters and body as Assert.

The standard json package does not escape the solidus, even though the RFC said so. Someone also report this as an error in RFC 3159 by removing solidus from list of escaped characters but the author itself reject it.

The panic is caused by the item in slice of Childs is being removed during iteration.

To fix this, we remove the childs on the second iteration after we remove any sub directories inside them.

Instead of calling filepath.EvalSymlink and Lstat, call os.Stat directly to the symlink system path.

This also fix the modTime not currently set to the original file when creating Node from symlink-ed file.

On container, sometimes the test fail with the following error

This was caused by server is not ready yet to accept connection when testing executed.

In the _doc, we remove generated HTML files.

In the index, we add link to README and section about Development that include links to repository, issues, and patches.

In the README, we reformat it to use AsciiDoc markup, remove the sanitizer library, add CLI for epoch, ini, and xtrk.

Previously, calling Merge(…​), set the merged MemFS Root.SysPath to "..". Since we allow the TryDirect to access the file directly (if its set to true), this may cause the file system leaks if returned MemFS set this flag to true.

To prevent that, we set the SysPath to the first MemFS SysPath.

This changes the usage of Development flag.

TryDirect define a flag to bypass file in memory. If its true, any call to Get will try direct read to file system.

This flag has several use cases. First, to test serving file system directly from disk during development. Second, to combine embedded MemFS instance with non-embedded instance. One is reading content from memory, one is reading content from disk directly.

The go-bench.sh accept two arguments: the method or function to benchmark, default to "."; and benchmark number, default to current timestamp YYYYmmDD-HHMM.

The script accept one single argument: the path to package to be tested. If its empty default to current directory and sub-directories.

The go-mod-tip shell script get and print the latest Go module version based on the last tag and the latest commit hash from the current working directory.

This command usually used to fix go.mod due to force commit.

The flag for RFC1123 comes with two options one with string timezone (-rfc1123) and one with numeric time zone (-rfc1123z).

This is the CLI that implements the lib/ini for getting and setting the key’s value from INI file.

Data contains predefined input and output values that is loaded from file to be used during test.

The data provides zero or more flags, an optional description, zero or more input, and zero or more output.

The data file name must end with ".txt".

The data content use the following format,

The data can contains zero or more flag. A flag is key and value separated by ":". The flag key must not contain spaces.

The data may contain description.

The line that start with "\n>>>" defined the beginning of input. An input can have a name, if its empty it will be set to "default". An input can be defined multiple times, with different names.

The line that start with "\n<<<" defined the beginning of output. An output can have a name, if its empty it will be set to "default". An output also can be defined multiple times, with different names.

Previously, if INI file contains multi line variables, for example

The Get and saved value is "a \tb", where it should be "a b" for Get and "a \\\n\t\b" again when saved.

This changes require refactoring how the variable’s value is parsed and stored. A variable value is parsed and stored from character after "=" until new line or comment as raw value, and the real value is derived by trimming white spaces, handle escaped character and double quotes.

If the field is pointer, the code will thrown panic if its point to nil struct or print "<invalid reflct.Value>" for String.

If node exist in memory, error on Update does not means the file is not exist. The node may have been embedded and then merged with other MemFS instance with Development flag set to true.

The SOA field defined the root authority for all zones and records served under the Server.

If the EnableIndexHtml in the ServeOptions enabled, server generate list of files inside the requested path as HTML.

A colon : is escaped using double backslash \\, for example a:b\\:c:d contains section a, subsection b:c, and variable d.

A double quote " is escaped using triple backslash, for example (\\\").

Since the sanitize package only contains HTML function, and the html package already exist, we move the function into html package.

The Close method flush and close all log forwarders. Any write to a closed MultiLogger will be ignored.

This changes require adding sync.Mutex to mark if the instance has been closed or not; which affect createMultiLogger and defaultMLog to return a pointer to prevent copy on Mutex.

The MarshalJSON method convert the Clise into slice by calling Slice and then convert it into JSON.

The Marshal function marshal the obj value to []byte by calling one of the method: MarshalBinary, MarshalJSON, or MarshalText; in respective order.

If obj implement one of the method with valid signature, it will return (out, nil, true); unless there is an error.

If the method signature invalid it will return (nil, err, false).

If obj is nil or none of the method exist it will return (nil, nil, false).

Given an input string, The NormalizeForID normalize it to HTML ID. The normalization follow Mozilla specification [1] rules,

The NormalizeForID do this normalization,

[1] https://developer.mozilla.org/en-US/docs/Web/HTML/Global_attributes/id.

UnmarshalForm read struct fields tagged with form: from out as key and set its using the value from url.Values based on that key. If the field does not have form: tag but it is exported, then it will use the field name, in case insensitive.

Only the following types are supported: bool, int/intX, uint/uintX, floatX, string, []byte, or type that implement BinaryUnmarshaler (UnmarshalBinary), json.Unmarshaler (UnmarshalJSON), or TextUnmarshaler (UnmarshalText).

A bool type can be set to true using the following string value: "true", "yes", or "1".

If the input contains multiple values but the field type is not slice, the field will be set using the first value.

It will return an error if the out variable is not set-able (the type is not a pointer to a struct). It will not return an error if one of the input value is not match with field type.

The Set function set the obj value by converting the string val from parameter to the obj type.

If the obj is an interface or struct, its value will be set by calling Unmarshal.

It will return an error if,

The Unmarshal function set the obj value by calling one of the method: UnmarshalBinary, UnmarshalJSON, or UnmarshalText; in respective order.

Just like reflect, the obj value must be pointer to initialized variable (&T) or pointer-to-pointer to uninitialized variable (**T).

If obj implement one of the method, it will return (true, nil) if there is no error.

If none of the method exist on obj, it will return (false, nil).

Given a StructField and the name of tag, return the tag’s value and options inside the tag. The options is any string after tag’s value, separated by comma. For example, given the following field definition

It will return (name, [opt1 opt2], true).

If the field is exported but does not have tag, it will return the field name (as is without converting to lower case) in val with hasTag is false: (Name, nil, false).

If the field is un-exported it will return empty val with hasTag is false ("", nil, false).

Replace ":=" with "var" and realign the field assignments.

Instead of using bytes.Replace, three times, iterate the plain text manually to clean up the white and multiple spaces.

Benchmark result,

Instead of using two bytes.Buffer pool, use one; and add space after time and prefix by writing to buffer directly instead of allocating new arguments to Fprintf.

Benchmark result,

Using Shutdown allow active connection not interrupted but it may cause delay when restarting the server.

While at it, set the doh and dot server instance to nil to release the resource, in case the Server need to start again.

Changes,

Plus other structs in the tests.