Re: Error installing matplotlib

On 9/3/25 07:20, Rob Cliffe wrote:
> 
> 
> On 03/09/2025 00:01, Mats Wichmann wrote:
>> On 9/2/25 14:51, Rob Cliffe via Python-list wrote:
>>>
>> There are two roots here:
>>
>> (1) it's not finding a prebuilt wheel.  You can see that because it's 
>> proposing to use the source distribution instead:
>>
>> > Collecting matplotlib
>> >    Using cached matplotlib-3.10.6.tar.gz (34.8 MB)
>>
>> and so it, in the initial output.
> I may be wrong, but I suspect that this is because I have attempted to 
> do the install multiple times.  I.e. it didn't say this on the first run.
>>
>> (2) Although at this point all hope of success is often lost on 
>> Windows, it's still going to try to compile things. It found gcc - 
>> there are messages from g++ in the output. Last I know, Pybind11 only 
>> builds with the Microsoft compiler because it uses msvc-specific types 
>> which gcc does not know about.
>>
>> Since the matplotlib project definition provides wheels for Python 
>> 3.13 on Windows (not just the main Python, but also the free-threaded 
>> versions), you need to figure out why the resolver doesn't think 
>> there's anything applicable available.
>>
>>
> Thanks for your reply, Mats, but I'm still lost.
> Best wishes
> Rob Cliffe


I can't tell you specifically how to diagnose why it's not finding what 
you hope it would find.  You might try these things:

# remove cached entries so you're resolving all from PyPI
pip cache purge

# do a test install with some flags that *might* give you useful info:
pip -v install --dry-run --only-binary :all: --ignore-installed matplotlib

It would take people with more pip experience to dig deeper.

I think the alternate manager "uv" may have more targeted debug control 
for this scenario, but I don't promise, haven't used it that much.

I'm sure somebody will pop up with comments if this is bad advice - I 
tried to answer the question of what's happening but not sure how to fix it.
-- 
https://mail.python.org/mailman3//lists/python-list.python.org