Update the debugging channels docu.

This commit is contained in:
Dimitrie O. Paun 2002-08-14 20:59:33 +00:00 committed by Alexandre Julliard
parent 3528755046
commit 8d8c709274

View file

@ -9,15 +9,6 @@
</para>
<note>
<para>
The new debugging interface can be considered to be
stable, with the exception of the in-memory message
construction functions. However, there is still a lot of
work to be done to polish things up. To make my life
easier, please follow the guidelines described in this
document.
</para>
<para>
It is possible to turn on and of debugging output from
within the debuger using the set command. Please see the
@ -26,13 +17,6 @@
</note>
<important>
<para>
Read this document before writing new code. DO NOT USE
<function>fprintf</function> (or
<function>printf</function>) to output things. Also, instead
of writing FIXMEs in the source, output a FIXME message if
you can.
</para>
<para>
At the end of the document, there is a "Style Guide" for
debugging messages. Please read it.
@ -43,16 +27,16 @@
<title>Debugging classes</title>
<para>
There are 4 types (or classes) of debugging messages:
There are 4 types (or classes) of messages:
</para>
<variablelist>
<varlistentry>
<term><literal>FIXME</literal></term>
<listitem>
<para>
Messages in this class relate to behavior of Wine that
does not correspond to standard Windows behavior and
that should be fixed.
Messages in this class are meant to signal unimplemented
features, known bugs, etc. They serve as a constant and
active reminder of what needs to be done.
</para>
<para>Examples: stubs, semi-implemented features, etc.</para>
</listitem>
@ -62,11 +46,9 @@
<listitem>
<para>
Messages in this class relate to serious errors in
Wine. This sort of messages are close to asserts --
that is, you should output an error message when the
code detects a condition which should not happen. In
other words, important things that are not warnings
(see below), are errors.
Wine. This sort of messages signal an inconsistent
internal state, or more general, a condition which
should never happen by design.
</para>
<para>
Examples: unexpected change in internal state, etc.
@ -86,8 +68,7 @@
not deal correctly with it, output a fixme.
</para>
<para>
Examples: fail to access a resource required by the
app, etc.
Examples: fail to access a resource required by the app.
</para>
</listitem>
</varlistentry>
@ -96,7 +77,7 @@
<listitem>
<para>
These are detailed debugging messages that are mainly
useful to debug a component. These are usually turned
useful to debug a component. These are usually turned
off.
</para>
<para>
@ -107,60 +88,14 @@
</listitem>
</varlistentry>
</variablelist>
<para>
The user has the capability to turn on or off messages of a
particular type. You can expect the following patterns of
usage (but note that any combination is possible):
</para>
<itemizedlist>
<listitem>
<para>
when you debug a component, all types
(<literal>TRACE</literal>, <literal>WARN</literal>,
<literal>ERR</literal>, <literal>FIXME</literal>) will
be enabled.
</para>
</listitem>
<listitem>
<para>
during the pre-alpha (maybe alpha) stage of Wine, most
likely the <literal>TRACE</literal> class will be
disabled by default, but all others
(<literal>WARN</literal>, <literal>ERR</literal>,
<literal>FIXME</literal>) will be enabled by default.
</para>
</listitem>
<listitem>
<para>
when Wine will become stable, most likely the
<literal>TRACE</literal> and <literal>WARN</literal>
classes will be disabled by default, but all
<literal>ERR</literal>s and <literal>FIXME</literal>s
will be enabled.
</para>
</listitem>
<listitem>
<para>
in some installations that want the smallest footprint
and where the debug information is of no interest, all
classes may be disabled by default.
</para>
</listitem>
</itemizedlist>
<para>
Of course, the user will have the runtime ability to
override these defaults. However, this ability may be turned
off and certain classes of messages may be completely
disabled at compile time to reduce the size of Wine.
</para>
</sect1>
<sect1 id="dbg-channels">
<title>Debugging channels</title>
<para>
Also, we divide the debugging messages on a component basis.
To better manage the large volume of debugging messages that
Wine can generate, we divide them also on a component basis.
Each component is assigned a debugging channel. The
identifier of the channel must be a valid C identifier but
note that it may also be a reserved word like
@ -177,109 +112,57 @@
<para>
We will refer to a generic channel as <literal>xxx</literal>.
</para>
<note>
<para>
for those who know the old interface, the channel/type is
what followed the _ in the
<function>dprintf_xxx</function> statements. For example,
to output a message on the debugging channel
<literal>reg</literal> in the old interface you would had
to write:
</para>
<programlisting>
dprintf_reg(stddeb, "Could not access key!\n");
</programlisting>
<para>
In the new interface, we drop the
<literal>stddeb</literal> as it is implicit. However, we
add an orthogonal piece of information to the message: its
class. This is very important as it will allow us to
selectively turn on or off certain messages based on the
type of information they report. For this reason it is
essential to choose the right class for the message.
Anyhow, suppose we figured that this message should belong
in the <literal>WARN</literal> class, so in the new
interface, you write:
</para>
<programlisting>
WARN(reg, "Could not access key!\n");
</programlisting>
</note>
</sect1>
<sect1 id="dbg-using">
<title>How to use it</title>
<para>
So, to output a message (class <literal>YYY</literal>) on
channel <literal>xxx</literal>, do:
</para>
Typically, a file contains code pertaining to only one component,
and as such, there is only one channel to output to. To simplify
usage, you can declare that channel at the beginning of the file,
and simply write FIXMEs, ERRs, etc. as such:
<programlisting>
#include "debugtools.h"
....
YYY(xxx, "&lt;message>", ...);
</programlisting>
<para>
Some examples from the code:
</para>
<programlisting>
#include "debugtools.h"
#include "wine/debug.h"
WINE_DEFAULT_DEBUG_CHANNEL(xxx);
...
TRACE(crtdll, "CRTDLL_setbuf(file %p buf %p)", file, buf);
WARN(aspi, "Error opening device errno=%d", save_error);
FIXME("some unimplemented feature", ...);
...
if (zero != 0)
ERR("This should never be non-null: %d", zero);
...
</programlisting>
</para>
<para>
I rare situations there is a need to output to more then one
debug channel per file. In such cases, you need to declare
all the additional channels at the top of the file, and
use the _-version of the debugging macros:
<programlisting>
#include "wine/debug.h"
WINE_DEFAULT_DEBUG_CHANNEL(xxx);
WINE_DECLARE_DEBUG_CHANNEL(yyy);
WINE_DECLARE_DEBUG_CHANNEL(zzz);
...
FIXME("this one goes to xxx channel");
...
FIXME_(yyy)("Some other msg for the yyy channel");
...
WARN_(zzz)("And yet another msg on another channel!");
...
</programlisting>
</para>
<para>
If you need to declare a new debugging channel, use it in
your code and then do:
</para>
<screen>
%tools/make_debug
</screen>
<para>
in the root directory of Wine. Note that this will result in
almost complete recompilation of Wine.
If you need to declare a new debugging channel, simply use it in
your code. It will be picked up automatically by the build process.
</para>
<note>
<orderedlist>
<listitem>
<para>
Please pay attention to which class you assign the
message. There are only 4 classes, so it is not hard.
The reason it is important to get it right is that too
much information is no information. For example, if
you put things into the <literal>WARN</literal> class
that should really be in the <literal>TRACE</literal>
class, the output will be too big and this will force
the user to turn warnings off. But this way he will
fail to see the important ones. Also, if you put
warnings into the <literal>TRACE</literal> class lets
say, he will most likely miss those because usually
the <literal>TRACE</literal> class is turned off. A
similar argument can be made if you mix any other two
classes.
</para>
</listitem>
<listitem>
<para>
All lines should end with a newline. If you can NOT
output everything that you want in the line with only
one statement, then you need to build the string in
memory. Please read the section below "In-memory
messages" on the preferred way to do it. PLEASE USE
THAT INTERFACE TO BUILD MESSAGES IN MEMORY. The reason
is that we are not sure that we like it and having
everything in one format will facilitate the
(automatic) translation to a better interface.
</para>
</listitem>
</orderedlist>
</note>
</sect1>
<sect1 id="dbg-checking">
@ -315,155 +198,6 @@ if(TRACE_ON(atom)){
</note>
</sect1>
<sect1 id="dbg-in-memory">
<title>In-memory messages</title>
<para>
If you NEED to build the message from multiple calls, you
need to build it in memory. To do that, you should use the
following interface:
</para>
<orderedlist>
<listitem>
<para>
declare a string (where you are allowed to declare C
variables) as follows:
<programlisting>
dbg_decl_str(name, len);
</programlisting>
where <parameter>name</parameter> is the name of the
string (you should use the channel name on which you
are going to output it)
</para>
</listitem>
<listitem>
<para>
print in it with:
<programlisting>
dsprintf(name, "&lt;message>", ...);
</programlisting>
which is just like a <function>sprintf</function>
function but instead of a C string as first parameter it
takes the name you used to declare it.
</para>
</listitem>
<listitem>
<para>
obtain a pointer to the string with: <function>dbg_str(name)</function>
</para>
</listitem>
<listitem>
<para>
reset the string (if you want to reuse it with):
<programlisting>
dbg_reset_str(name);
</programlisting>
</para>
</listitem>
</orderedlist>
<para>
Example (modified from the code):
</para>
<programlisting>
void some_func(tabs)
{
INT32 i;
LPINT16 p = (LPINT16)tabs;
dbg_decl_str(listbox, 256); /* declare the string */
for (i = 0; i &lt; descr->nb_tabs; i++) {
descr->tabs[i] = *p++&lt;&lt;1;
if(TRACING(listbox)) /* write in it only if
dsprintf(listbox, "%hd ", descr->tabs[i]); /* we are gonna output it */
}
TRACE(listbox, "Listbox %04x: settabstops %s",
wnd->hwndSelf, dbg_str(listbox)); /* output the whole thing */
}
</programlisting>
<para>
If you need to use it two times in the same scope do like
this:
</para>
<programlisting>
void some_func(tabs)
{
INT32 i;
LPINT16 p = (LPINT16)tabs;
dbg_decl_str(listbox, 256); /* declare the string */
for (i = 0; i &lt; descr->nb_tabs; i++) {
descr->tabs[i] = *p++&lt;&lt;1;
if(TRACING(listbox)) /* write in it only if
dsprintf(listbox, "%hd ", descr->tabs[i]); /* we are gonna output it */
}
TRACE(listbox, "Listbox %04x: settabstops %s\n",
wnd->hwndSelf, dbg_str(listbox)); /* output the whole thing */
dbg_reset_str(listbox); /* !!!reset the string!!! */
for (i = 0; i &lt; descr->extrainfo_nr; i++) {
descr->extrainfo = *p+1;
if(TRACING(listbox)) /* write in it only if
dsprintf(listbox,"%3d ",descr->extrainfo); /* we are gonna output it */
}
TRACE(listbox, "Listbox %04x: extrainfo %s\n",
wnd->hwndSelf, dbg_str(listbox)); /* output the whole thing */
}
</programlisting>
<important>
<para>
As I already stated, I do not think this will be the
ultimate interface for building in-memory debugging
messages. In fact, I do have better ideas which I hope to
have time to implement for the next release. For this
reason, please try not to use it. However, if you need to
output a line in more than one
<function>dprintf_xxx</function> calls, then USE THIS
INTERFACE. DO NOT use other methods. This way, I will
easily translate everything to the new interface (when it
will become available). So, if you need to use it, then
follow the following guidelines:
</para>
<itemizedlist>
<listitem>
<para>wrap calls to <function>dsprintf</function> with a
</para>
<programlisting>
if(YYY(xxx))
dsprintf(xxx,...);
</programlisting>
<para>
Of course, if the call to
<function>dsprintf</function> is made from within a
function which you know is called only if
<function>YYY(xxx)</function> is true, for example if
you call it only like this:
</para>
<programlisting>
if(YYY(xxx))
print_some_debug_info();
</programlisting>
<para>
then you need not (and should not) wrap calls to
<function>dsprintf</function> with the before
mentioned <function>if</function>.
</para>
</listitem>
<listitem>
<para>
name the string EXACTLY like the debugging channel on
which is going to be output. Please see the above
example.
</para>
</listitem>
</itemizedlist>
</important>
</sect1>
<sect1 id="dbg-resource-ids">
<title>Resource identifiers</title>
@ -474,7 +208,7 @@ if(YYY(xxx))
introduced a new function called <function>debugres</function>.
</para>
<para>
The function is defined in <filename>debugstr.h</filename>
The function is defined in <filename>wine/debug.h</filename>
and has the following prototype:
</para>
<programlisting>
@ -482,35 +216,28 @@ LPSTR debugres(const void *id);
</programlisting>
<para>
It takes a pointer to the resource id and returns a nicely
formatted string of the identifier. If the high word of the
pointer is <literal>0</literal>, then it assumes that the
identifier is a number and thus returns a string of the
form:
formatted string of the identifier (which can be a string or
a number, depending on the value of the high word).
Numbers are formatted as such:
</para>
<programlisting>
#xxxx
</programlisting>
<para>
where <literal>xxxx</literal> are 4 hex-digits representing
the low word of <parameter>id</parameter>.
</para>
<para>
If the high word of the pointer is not <literal>0</literal>,
then it assumes that the identifier is a string and thus
returns a string of the form:
while strings as:
</para>
<programlisting>
'&lt;identifier>'
'some-string'
</programlisting>
<para>
Thus, to use it, do something on the following lines:
Simply use it in your code like this:
</para>
<programlisting>
#include "debugtools.h"
#include "wine/debug.h"
...
YYY(xxx, "resource is %s", debugres(myresource));
TRACE("resource is %s", debugres(myresource));
</programlisting>
</sect1>
@ -518,14 +245,15 @@ LPSTR debugres(const void *id);
<title>The <parameter>--debugmsg</parameter> command line option</title>
<para>
So, the <parameter>--debugmsg</parameter> command line
option has been changed as follows:
The <parameter>--debugmsg</parameter> command line
option controls the output of the debug messages.
It has the following syntax:
<parameter>--debugmsg [yyy]#xxx[,[yyy1]#xxx1]*</parameter>
</para>
<itemizedlist>
<listitem>
<para>
the new syntax is: <parameter>--debugmsg
[yyy]#xxx[,[yyy1]#xxx1]*</parameter> where
where
<literal>#</literal> is either <literal>+</literal> or
<literal>-</literal>
</para>
@ -544,8 +272,7 @@ LPSTR debugres(const void *id);
<para>
enables all messages on the <literal>reg</literal>
channel and disables all messages on the
<literal>file</literal> channel. This is same as the old
semantics.
<literal>file</literal> channel.
</para>
</listitem>
<listitem>
@ -660,7 +387,7 @@ where:
it as the first thing and include them in parentheses,
like this:
<programlisting>
YYY(xxx, "(%d,%p,etc)...\n", par1, par2, ...);
TRACE("(%d, %p, ...)\n", par1, par2, ...);
</programlisting>
</para>
</listitem>
@ -669,17 +396,14 @@ YYY(xxx, "(%d,%p,etc)...\n", par1, par2, ...);
for stubs, you should output a <literal>FIXME</literal>
message. I suggest this style:
<programlisting>
FIXME(xxx, "(%x,%d...): stub\n", par1, par2, ...);
FIXME("(%x, %d, ...): stub\n", par1, par2, ...);
</programlisting>
That is, you output the parameters, then a : and then a string
containing the word "stub". I've seen "empty stub", and others, but I
think that just "stub" suffices.
</para>
</listitem>
<listitem>
<para>
output 1 and ONLY 1 line per message. That is, the format
string should contain only 1 <literal>\n</literal> and it
try to output one line per message. That is, the format
string should contain only one <literal>\n</literal> and it
should always appear at the end of the string. (there are
many reasons for this requirement, one of them is that
each debug macro adds things to the beginning of the line)
@ -690,11 +414,11 @@ YYY(xxx, "(%d,%p,etc)...\n", par1, par2, ...);
if you want to name a value, use <literal>=</literal> and
NOT <literal>:</literal>. That is, instead of saying:
<programlisting>
FIXME(xxx, "(fd: %d, file: %s): stub\n", fd, name);
FIXME("(fd: %d, file: %s): stub\n", fd, name);
</programlisting>
say:
<programlisting>
FIXME(xxx, "(fd=%d, file=%s): stub\n", fd, name);
FIXME("(fd=%d, file=%s): stub\n", fd, name);
</programlisting>
use <literal>:</literal> to separate categories.
</para>
@ -705,7 +429,7 @@ FIXME(xxx, "(fd=%d, file=%s): stub\n", fd, name);
<programlisting>
FIXME(xxx, "(fd=%d, file=%s)\n", fd, name);
</programlisting>
but use:
instead use:
<programlisting>
FIXME(xxx, "(fd=%d, file=%s): stub\n", fd, name);
</programlisting>
@ -727,7 +451,7 @@ FIXME(xxx, "(fd=%d, file=%s): stub\n", fd, name);
<programlisting>
HANDLE32 WINAPI YourFunc(LPCSTR s)
{
FIXME(xxx, "(%s): stub\n", debugstr_a(s));
FIXME("(%s): stub\n", debugstr_a(s));
}
</programlisting>
</para>
@ -739,7 +463,7 @@ HANDLE32 WINAPI YourFunc(LPCSTR s)
<programlisting>
HANDLE32 WINAPI YourFunc(LPCSTR res)
{
FIXME(xxx, "(res=%s): stub\n", debugres(s));
FIXME("(res=%s): stub\n", debugres(s));
}
</programlisting>
if the resource identifier is a <type>SEGPTR</type>, use