This class is a collection of utility methods and attributes, mainly around memory management. To simplify and speed up its use, it is automatically instantiated under the name TOOLS when PyMuPDF is imported.

Method / Attribute Description
Tools.gen_id() generate a unique identifyer
Tools.image_profile() report basic image properties
Tools.store_shrink() shrink the storables cache [1]
Tools.mupdf_warnings() return the accumulated MuPDF warnings
Tools.mupdf_display_errors() return the accumulated MuPDF warnings
Tools.reset_mupdf_warnings() empty MuPDF messages on STDOUT
Tools.set_aa_level() set the anti-aliasing values
Tools.set_annot_stem() set the prefix of new annotation / link ids
Tools.set_small_glyph_heights() search and extract using small bbox heights
Tools.set_subset_fontnames() control suppression of subset fontname tags
Tools.show_aa_level() return the anti-aliasing values
Tools.unset_quad_corrections() disable PyMuPDF-specific code
Tools.fitz_config configuration settings of PyMuPDF
Tools.store_maxsize maximum storables cache size
Tools.store_size current storables cache size

Class API

class Tools

A convenience method returning a unique positive integer which will increase by 1 on every invocation. Example usages include creating unique keys in databases - its creation should be faster than using timestamps by an order of magnitude.


MuPDF has dropped support for this in v1.14.0, so we have re-implemented a similar function with the following differences:

  • It is not part of MuPDF’s global context and not threadsafe (not an issue because we do not support threads in PyMuPDF anyway).
  • It is implemented as int. This means that the maximum number is sys.maxsize. Should this number ever be exceeded, the counter starts over again at 1.
Return type:int
Returns:a unique positive integer.

(New in v1.18.6)

Set or inquire the prefix for the id of new annotations, fields or links.

Parameters:stem (str) – if omitted, the current value is returned, default is “fitz”. Annotations, fields / widgets and links technically are subtypes of the same type of object (/Annot) in PDF documents. An /Annot object may be given a unique identifier within a page. For each of the applicable subtypes, PyMuPDF generates identifiers “stem-Annn”, “stem-Wnnn” or “stem-Lnnn” respectively. The number “nnn” is used to enforce the required uniqueness.
Return type:str
Returns:the current value.

(New in v1.18.5)

Set or inquire reduced bbox heights in text extract and text search methods.

Parameters:on (bool) – if omitted or None, the current setting is returned. For other values the bool() function is applied to set a global variable. If True, Page.search_for() and Page.get_text() methods return character, span, line or block bboxes that have a height of font size. If False (standard setting when PyMuPDF is imported), bbox height will be based on font properties and normally equal line height.
Return type:bool
Returns:True or False.


Text extraction options “xml”, “xhtml” and “html”, which directly wrap MuPDF code, are not influenced by this.


(New in v1.18.9)

Control suppression of subset fontname tags in text extractions.

Parameters:on (bool) – if omitted / None, the current setting is returned. Arguments evaluating to True or False set a global variable. If True, options “dict”, “json”, “rawdict” and “rawjson” will return e.g. "NOHSJV+Calibri-Light", otherwise only "Calibri-Light" (the default). The setting remains in effect until changed again.
Return type:bool
Returns:True or False.


Except mentioned above, no other text extraction variants are influenced by this. This is especially true for the options “xml”, “xhtml” and “html”, which are based on MuPDF code. They extract the font name "Calibri-Light", or even just the family name – Calibri in this example.


(New in v1.18.10)

Enable / disable PyMuPDF-specific code, that tries to rebuild valid character quads when encountering nonsense in Page.get_text() text extractions. This code depends on certain font properties (ascender and descender), which do not exist in rare situations and cause segmentation faults when trying to access them. This method sets a global parameter in PyMuPDF, which suppresses execution of this code.

Parameters:on (bool) – if omitted or None, the current setting is returned. For other values the bool() function is applied to set a global variable. If True, PyMuPDF will not try to access the resp. font properties and use values ascender=0.8 and descender=-0.2 instead.
Return type:bool
Returns:True or False.

(New in v1.16.17) Show important properties of an image provided as a memory area. Its main purpose is to avoid using other Python packages just to determine basic properties.

Parameters:stream (bytes,bytearray) – the image data.
Return type:dict
Returns:a dictionary with the keys “width”, “height”, “xres”, “yres”, “colorspace” (the colorspace.n value, number of colorants), “cs-name” (the value), “bpc”, “ext” (image type as file extension). The values for these keys are the same as returned by Document.extract_image(). Please also have a look at resolution.


  • For some “exotic” images (FAX encodings, RAW formats and the like), this method will not work and return None. You can however still work with such images in PyMuPDF, e.g. by using Document.extract_image() or create pixmaps via Pixmap(doc, xref). These methods will automatically convert exotic images to the PNG format before returning results.

  • Some examples:

    In [1]: import fitz
    In [2]: stream = open(<image.file>, "rb").read()
    In [3]: fitz.TOOLS.image_profile(stream)
    {'width': 439,
    'height': 501,
    'xres': 96,
    'yres': 96,
    'colorspace': 3,
    'bpc': 8,
    'ext': 'jpeg',
    'cs-name': 'DeviceRGB'}
    In [4]:<input.pdf>)
    In [5]: stream = doc.xref_stream_raw(5)  # no decompression!
    In [6]: fitz.TOOLS.image_profile(stream)
    {'width': 816,
    'height': 1056,
    'xres': 96,
    'yres': 96,
    'colorspace': 1,
    'bpc': 8,
    'ext': 'jpeg',
    'cs-name': 'DeviceGray'}

Reduce the storables cache by a percentage of its current size.

Parameters:percent (int) – the percentage of current size to free. If 100+ the store will be emptied, if zero, nothing will happen. MuPDF’s caching strategy is “least recently used”, so low-usage elements get deleted first.
Return type:int
Returns:the new current store size. Depending on the situation, the size reduction may be larger than the requested percentage.

(New in version 1.16.14) Return the current anti-aliasing values. These values control the rendering quality of graphics and text elements.

Return type:dict
Returns:A dictionary with the following initial content: {'graphics': 8, 'text': 8, 'graphics_min_line_width': 0.0}.

(New in version 1.16.14) Set the new number of bits to use for anti-aliasing. The same value is taken currently for graphics and text rendering. This might change in a future MuPDF release.

Parameters:level (int) – an integer ranging between 0 and 8. Value outside this range will be silently changed to valid values. The value will remain in effect throughout the current session or until changed again.

(New in version 1.16.0)

Empty MuPDF warnings message buffer.


(New in version 1.16.8)

Show or set whether MuPDF errors should be displayed.

Parameters:value (bool) – if not a bool, the current setting is returned. If true, MuPDF errors will be shown on sys.stderr, otherwise suppressed. In any case, messages continue to be stored in the warnings store. Upon import of PyMuPDF this value is True.
Returns:True or False

(New in version 1.16.0)

Return all stored MuPDF messages as a string with interspersed line-breaks.

Parameters:reset (bool) – (new in version 1.16.7) whether to automatically empty the store.

A dictionary containing the actual values used for configuring PyMuPDF and MuPDF. Also refer to the installation chapter. This is an overview of the keys, each of which describes the status of a support aspect.

Key Support included for …
plotter-g Gray colorspace rendering
plotter-rgb RGB colorspace rendering
plotter-cmyk CMYK colorspcae rendering
plotter-n overprint rendering
pdf PDF documents
xps XPS documents
svg SVG documents
cbz CBZ documents
img IMG documents
html HTML documents
epub EPUB documents
jpx JPEG2000 images
js JavaScript
tofu all TOFU fonts
tofu-cjk CJK font subset (China, Japan, Korea)
tofu-cjk-ext CJK font extensions
tofu-cjk-lang CJK font language extensions
tofu-emoji TOFU emoji fonts
tofu-historic TOFU historic fonts
tofu-symbol TOFU symbol fonts
tofu-sil TOFU SIL fonts
icc ICC profiles
py-memory using Python memory management [2]
base14 Base-14 fonts (should always be true)

For an explanation of the term “TOFU” see this Wikipedia article.:

In [1]: import fitz
In [2]: TOOLS.fitz_config
{'plotter-g': True,
 'plotter-rgb': True,
 'plotter-cmyk': True,
 'plotter-n': True,
 'pdf': True,
 'xps': True,
 'svg': True,
 'cbz': True,
 'img': True,
 'html': True,
 'epub': True,
 'jpx': True,
 'js': True,
 'tofu': False,
 'tofu-cjk': True,
 'tofu-cjk-ext': False,
 'tofu-cjk-lang': False,
 'tofu-emoji': False,
 'tofu-historic': False,
 'tofu-symbol': False,
 'tofu-sil': False,
 'icc': True,
 'py-memory': True, # (False if Python 2)
 'base14': True}
Return type:dict

Maximum storables cache size in bytes. PyMuPDF is generated with a value of 268’435’456 (256 MB, the default value), which you should therefore always see here. If this value is zero, then an “unlimited” growth is permitted.

Return type:int

Current storables cache size in bytes. This value may change (and will usually increase) with every use of a PyMuPDF function. It will (automatically) decrease only when Tools.store_maxize is going to be exceeded: in this case, MuPDF will evict low-usage objects until the value is again in range.

Return type:int

Example Session

>>> import fitz
# print the maximum and current cache sizes
>>> fitz.TOOLS.store_maxsize
>>> fitz.TOOLS.store_size
>>> doc ="demo1.pdf")
# pixmap creation puts lots of object in cache (text, images, fonts),
# apart from the pixmap itself
>>> pix = doc[0].get_pixmap(alpha=False)
>>> fitz.TOOLS.store_size
# release (at least) 50% of the storage
>>> fitz.TOOLS.store_shrink(50)
>>> fitz.TOOLS.store_size
# get a few unique numbers
>>> fitz.TOOLS.gen_id()
>>> fitz.TOOLS.gen_id()
>>> fitz.TOOLS.gen_id()
# close document and see how much cache is still in use
>>> doc.close()
>>> fitz.TOOLS.store_size


[1]This memory area is internally used by MuPDF, and it serves as a cache for objects that have already been read and interpreted, thus improving performance. The most bulky object types are images and also fonts. When an application starts up the MuPDF library (in our case this happens as part of import fitz), it must specify a maximum size for this area. PyMuPDF’s uses the default value (256 MB) to limit memory consumption. Use the methods here to control or investigate store usage. For example: even after a document has been closed and all related objects have been deleted, the store usage may still not drop down to zero. So you might want to enforce that before opening another document.
[2]Optionally, all dynamic management of memory can be done using Python C-level calls. MuPDF offers a hook to insert user-preferred memory managers. We are using option this for Python version 3 since PyMuPDF v1.13.19. At the same time, all memory allocation in PyMuPDF itself is also routed to Python (i.e. no more direct malloc() calls in the code). We have seen improved memory usage and slightly reduced runtimes with this option set. If you want to change this, you can set #define JM_MEMORY 0 (uses standard C malloc, or 1 for Python allocation )in file fitz.i and then generate PyMuPDF.