The Magic Number Database

How does it work?

MagNumDB is a database that contains about 380,000 items. These items are constants, names, values all extracted from more than 6,000 header files (.h, .hxx, .hpp, .idl, etc.) provided by standard Windows and Visual Studio SDKs and WDKs.

Some values have been extracted from the very special uuid.lib file that contains the value of thousands of guids and property keys, not present anywhere else in header files. It also contains around 30,0000 undocumented guids that I found in en envelope someone put in front of my door on a cold and dark night of april 2018.

"C2P5"

To build this database, we have tried many existing parsers, things like CLANG or other fine tools, but they just don't suit our needs. They can't handle tens of thousands of files (or maybe with 128G of RAM?), they can't handle some specific (or just very old) Microsoft constructs, they don't remember the stack of #define directives that led to a definition, they only give you a final AST, not a partial one, etc.

So, in the end, we have written a C/C++ parser named C2P5 (for C/CPP/PreProcessor/Parser), tailored specifically for computing constants. C2P5 is capable of preprocessing, parsing and partially evaluating all header files as if they were included in a one big virtual project (that of course, does not compile) on a 32G RAM machine. It currently supports the following preprocessor and C/C++ constructs:

#define preprocessor that define constants
#define preprocessor that define expression to compute constants
C and C++ enum definitions
Special GUID constructs like DECLSPEC_UUID, DEFINE_GUIDSTRUCT, MIDL_INTERFACE, you name it
Some IDL constructs
Etc.

The parser remembers dynamic preprocessor definitions (#if, #ifdef, etc.) that are conditions for constants definitions and expression computation. All parsed items are saved in the database, as well as the associated conditions. There may be more than one item corresponding to a given name, if there are differences in their associated conditions stack.

C2P5 supports the following types of constants, regardless of the way they are defined in source files:

integer, signed or unsigned, from 8-bit to 128-bit
floating point number, single (32-bit) or double (64-bit)
globally unique identifier, or guid, uuid, CLSID, IID, etc.
string, in narrow, wide, or utf-8/16/32 formats
character, in narrow, wide, or utf-8/16/32 formats
pointer, a constant cast as a pointer
property key, or PROPERTYKEY, or PKEY, a combination of a guid (fmtid) and a 32-bit integer (id).

C2P5 and this MagNumDB web site are written in C# and use a Lucene database as a full-text search engine. C2P5 uses a custom ANTLR4cs C grammar for expression parsing, not for preprocessor parsing.

Frequently Asked Questions

Q: Can I run custom queries?
A: Yes. You can use Lucene's query syntax. Note the Database is case insensitive. Check out the list of columns with their corresponding database column names:
- Index: the item's index.
- Name: the item's name, or title. For enum values, it's prefixed with the enum's name.
  Corresponding database field: title.
- Type: the item's type, expressed in a specific, universal system.
  Corresponding Lucene's field: valuetype, but use the value of the Storage type column for queries on items type.
- Value: the item's value. For string values, it's displayed as is w/o quotes nor escaping.
  Corresponding database field: value. There is also a typedvalue field that can contain Lucene's Int32, Int64, Float, Double or String typed values.
- Hex Value: the item's hexadecimal value, only valid for integer types.
  There is no corresponding database field.
- Signed Value: the item's signed value, only valid for integer and double types. Useful when you found a unsigned number and want to get the corresponding signed value (uint -> int for example).
  There is no corresponding database field.
- Source lines(s): the source lines from which the item was extracted. Comments and characters such as TAB, CR are preserved.
  Corresponding database field: source.
- File path: the file path from which the item was extracted, terminated by the line index.
  Corresponding database field: filepath. There are also directory and filename fields. filename is useful to query for values in a given file.
- Score: Lucene's score value. Items are returned sorted by score.
- Guid formats: a utility columns that proposes all possible guid formats. Only valid for guid types.
  There is no corresponding database field
- Char type: the character encoding (narrow, wide, etc.). Only valid for char and string values.
  Corresponding database field: chartype.
- Storage type: the exact storage type in the database. Only used for specific queries.
  Corresponding database field: valuetype.
- Conditions: the list of preprocessor conditions for a given item. If a condition is surrounded with a REVERSE function, it means the reverse condition of what's surrounded was the condition.
  Corresponding database field: conditions. There may be multiple conditions fields for one item.
Some important points to note:
- If you search for a negative number (like error codes expressed using signed integers), you can surround the search with double quotes, for example searching for "-2147009290" is equivalent to searching for 0x80073CF6, 80073CF6 (or even 3CF6 in this specific case). If you don't surround negative numbers with double quotes, the system will search for signed and unsigned corresponding values.
- The full-text search uses items' name, value, hex value, guid formats, source lines, etc. so refine your search if you get back too many items.
- You can search for guid binary values using just a part of their full definition. For example, if you search for '79eac9c5', it will get you 3 guid entries
- Leading wild card search (*whatever) is allowed, but it may slow down the search significantly.
Here are some example or custom queries:

title:wm_user returns the WM_USER Windows message item, not all items that reference the WM_USER token.

title:wm_u* returns all items (Windows messages probably) whose name starts with WM_U*.

value:1024 AND title:wm_* returns all items (Windows messages probably) whose name starts with WM_U* and value is 1024. Note AND must be UPPERCASE for Lucene's to understand it as an AND operator.

value:"00000002-0000-0000-C000-000000000046" returns the IMalloc IID guid value.
Q: I always get back 400 items at most. Why?
A: The MagNumDb site currently limits the number of returned items to this value by design. This should be enough for most real search scenarios (like enums with a big list of values).
Q: Some values are empty. Why?
A: The expression evaluator cannot always compute the final value, because it sometimes require contextual information (custom defines, compiler architecture, etc.) that the parser doesn't have. You can enable the Source line(s) column to check what's the corresponding source and try to work it out by yourself. The origin file path and line number is also included.
Q: Can I get all values for a given enum?
A: Sure. For example, __VSHPROPID is the name of an enum, and a query like this: parent:__VSHPROPID will get all the enum's values (66 entries).
Q: I sometimes get back items whose name starts with "__magnumdb__enum_". Why?
A: C allows unnamed enums. The MagNumDb parser just gives them a name, using the defining file name and a counter.
Q: I sometimes get back the same item/name twice or more with different values. For example: TARGET_IS_NT60_OR_LATER. Why?
A: This can be the case with defines. Their value may be different if they were surrounded by # directives (#if, #else, etc.) in the original source code. You can enable the Condition(s) column to check what are the corresponding conditions for a given item.
Note: if a condition is surrounded with a funny REVERSE function token, it means the reverse condition of what's surrounded.

Database Metrics

Last update date: Thursday December 13, 2018
Items count: 385322
Parsed files count: 6272
Windows SDK version: 10.0.17763.0

Contact

We welcome feedback.
Seen anything missing? A bug? A wrong value? Do you have any suggestion for improvements? Do you have an idea for a cool new feature?

Please contact us here

About MagNumDB

All product names, logos, and brands are property of their respective owners. All company, product and service names used in this website are for identification purposes only.

All values, names, source code fragments displayed here have been extracted from files that are property of their respective owners.

THIS SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND OTHER THAN AS SPECIFICALLY SET FORTH IN THE LICENSE AGREEMENT, INCLUDING WITHOUT LIMITATION WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Index	Score	Name	Type	Value	Hex value	Signed value	Guid formats	Char type	Storage type	Source line(s)	File path	Condition(s)
Index	Score	Name	Type	Value	Hex value	Signed value	Guid formats	Char type	Storage type	Source line(s)	File path	Condition(s)
1	4.236707	__MIDL___MIDL_itf_dxva2api_0000_0000_0012	Enum							enum __MIDL___MIDL_itf_dxva2api_0000_0000_0012 { DXVA2_PictureParametersBufferType = 0, DXVA2_MacroBlockControlBufferType = 1, DXVA2_ResidualDifferenceBufferType = 2, DXVA2_DeblockingControlBufferType = 3, DXVA2_InverseQuantizationMatrixBufferType = 4, DXVA2_SliceControlBufferType = 5, DXVA2_BitStreamDateBufferType = 6, DXVA2_MotionVectorBuffer = 7, DXVA2_FilmGrainBuffer = 8 } ;	%ProgramFiles(x86)%\Windows Kits\10\Include\10.0.17763.0\um\dxva2api.h(542,0)	If WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)